Techniques to Improve Model Output (20%)

Sampling Parameters

KEY CONCEPTS

  • Temperature: Controls randomness (0=deterministic, higher=creative)
  • Top-P (nucleus sampling): Probability-based token selection
  • Top-K: Fixed number of token candidates
  • Token count and output length limits
  • Safety settings for content filtering

WHAT THE EXAM IS REALLY TESTING

Know what each parameter does. Temperature is most commonly tested.

COMMON TRAPS

  • !Thinking temperature affects quality, not randomness
  • !Confusing Top-P and Top-K effects
  • !Forgetting safety settings exist

OFFICIAL DOCUMENTATION

PRACTICE QUESTIONS

3 questions available for this topic

PRACTICE NOW →