Practical Tuning Guide: Which Parameters to Change for Which Use Case

~15 min read

A concrete parameter cheat sheet across 3 common use cases — creative writing, factual QA, and code generation — showing which of the 7 levers actually matter for each, and in which direction to tune them.

Practical Tuning Guide: Which Parameters to Change for Which Use Case is a Pro topic

Sign in, then upgrade to Pro or Power to unlock this topic and the full AI Engineering curriculum.

Key points

  • Creative writing: high temperature (0.8-1.0), generous top-p, moderate presence penalty, multinomial sampling
  • Factual QA/chatbots: low temperature (0.0-0.3), tight top-p, near-zero penalties, greedy-ish decoding
  • Code generation: low temperature like factual QA, but with tight stop sequences and generous max tokens to avoid truncated code
  • Temperature + sampling strategy is the primary diversity dial; penalties are secondary fine-tuning on top of that
  • Stop sequences and max tokens are about boundary control, not creativity — get these right regardless of use case