AI Ultimate AI Engineer 4

Question 1

Which serving optimization packs multiple independent requests into a single GPU forward pass without waiting for a full batch to fill?

Accepted Answer

Continuous batching (iteration-level scheduling)

Answer

Continuous batching inserts new requests mid-generation at the token level, maximizing GPU utilization without fixed batch wait times.

Question 2

An AI engineer wants to reduce an LLM's memory footprint by 4x with minimal accuracy loss. Which technique is most appropriate?

Accepted Answer

INT4 weight quantization

Answer

INT4 quantization represents weights in 4 bits instead of 16, achieving approximately 4x memory reduction with small accuracy degradation.

Question 3

What is the key advantage of Flash Attention over standard attention implementation?

Accepted Answer

It computes attention in tiles to avoid materializing the full N×N attention matrix in HBM

Answer

Flash Attention tiles the computation to keep intermediates in SRAM rather than writing the full attention matrix to GPU high-bandwidth memory, dramatically reducing memory I/O.

Question 4

Which technique enables an AI system to use external tools like calculators or web search during inference?

Accepted Answer

Function calling / tool use via structured output parsing

Answer

Function calling allows the model to emit structured JSON specifying a tool and arguments, which the host application executes and returns results for.

Question 5

In MLOps, what does 'model drift' refer to?

Accepted Answer

Degradation in model performance caused by changes in real-world data distribution

Answer

Model drift occurs when the statistical properties of production inputs diverge from the training distribution, causing accuracy to degrade over time.

AI - Engineer Practice Test

AI - Engineer Practice Test

AI Ultimate AI Engineer 4