Introduction to AI Models on GPU quiz

1. What is the key difference between training and inference?

   Training uses only forward pass; inference uses forward + backward

   Training updates weights; inference does not update weights

   Inference requires larger datasets than training

   Inference always uses CPUs

2. What do model formats primarily define?

   The compression ratio of training logs

   How weights, biases, and metadata are serialized

   The number of tokens a model can generate

   How GPUs are scheduled

3. TensorRT engines are best described as:

   Portable checkpoints for research

   Vendor-agnostic exchange formats

   GPU-optimized binaries compiled for specific hardware

   Lightweight community-driven formats

4. What makes GPUs well-suited for inference compared to CPUs?

   They use fewer cores

   They have thousands of parallel cores optimized for matrix math

   They consume a lot of power per core

   They handle more diverse instructions

5. What is the “knee point” in performance optimization?

   The value of latency where throughput collapses

   The point where throughput improves without hurting latency too much

   The slowest observed latency

   The minimal memory footprint achievable for a model, regardless of performance trade-offs

error_outline An error occurred when grading the quiz. Please try again.