Triton Inference Server Quiz

Answers and explanations appear as you take the quiz.

1. What makes Triton Inference Server 'framework-agnostic'?
   • A single instance can serve models from PyTorch, TensorFlow, ONNX, TensorRT, and custom backends at once
   • It only runs TensorRT engines
   • It converts every model into Java
   • It requires all models to be retrained in one framework
2. How does dynamic batching improve performance?
   • By processing every request the instant it arrives, one at a time
   • By grouping requests that arrive close together into a single batch for the GPU
   • By deleting requests that arrive too slowly
   • By moving inference to the CPU
3. What trade-off can Triton dynamic batching introduce?
   • Every request becomes lower latency regardless of traffic
   • It raises model accuracy by changing the weights
   • Compatible stateless requests may gain throughput while queued requests wait within a configured delay
   • It only works when inference runs on a CPU
4. What does a Triton 'model ensemble' (or business-logic scripting) let you do?
   • Train several models together
   • Encrypt the model files
   • Reduce the model's accuracy
   • Chain preprocessing, inference, and postprocessing into one server-side pipeline
5. What does Triton concurrent model execution permit?
   • Multiple models or instances to execute in parallel on the same system, subject to available resources
   • Only one model may run until the server restarts
   • A model to change its weights during each request
   • Requests to bypass the scheduler and share another model output
6. Which protocols does Triton commonly expose for clients to send inference requests?
   • Only FTP
   • HTTP/REST and gRPC
   • Only email
   • Only Bluetooth
7. Which monitoring interface does the Triton guide say it exposes?
   • A built-in SQL reporting dashboard
   • A browser-only interactive notebook
   • Prometheus metrics
   • A mandatory cloud billing API
8. What should a serving team measure when tuning dynamic batching?
   • Only the largest possible batch size
   • Only average model parameter count
   • Only startup time on an idle server
   • Throughput and latency under representative request traffic