Bitzu

AI-901 ยท AI models and how to configure them

Generation parameters that shape output

Generation settings that change output: temperature, top_p, max output tokens, stop sequences, frequency and presence penalties, and when to adjust each.

Key points

  • Temperature: rescales token probabilities; low: focused, repeatable; high: varied, creative.
  • Top_p (nucleus sampling): keeps the most probable tokens summing to p.
  • Max output tokens: caps length; too small cuts answers mid-sentence.
  • Stop sequence: ends generation when that text appears.
  • Frequency penalty: discourages tokens by how often they appeared.
  • Presence penalty: discourages any token that appeared, nudging toward new topics.

Practice questions

Single choice

What is the general effect of lowering the temperature?

  1. Output becomes more focused and more repeatable
  2. Output becomes longer and more creative
  3. The context window gets larger
  4. The model switches to a cheaper pricing tier
Show answer

Answer: A. Output becomes more focused and more repeatable

A lower temperature concentrates probability on the most likely tokens, so answers become more focused and repeatable. Raising it adds variety. It does not change response length or the context window.

Select all that apply

Which settings help reduce repetitive wording in a response?

  1. Frequency penalty
  2. Presence penalty
  3. Max output tokens
  4. Stop sequence
  5. Context window size
Show answer

Answer: A. Frequency penalty, B. Presence penalty

Frequency and presence penalties both discourage reusing tokens that already appeared. Max output tokens only cuts length, a stop sequence only ends generation, and the context window is not a tuning knob for style.

True or false

True or false: setting max output tokens to a small number makes the model write a shorter but complete answer.

Show answer

Answer: False

False. The cap does not tell the model to be concise; generation is simply cut off when the limit is reached, so the answer can end mid-sentence. To get shorter answers, ask for brevity in the prompt.

Practice 8 questions on this topic in Bitzu

Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.

Download on the App StoreGet it on Google Play

More AI-901 topics in this module