what do 'temperature' and 'top-p' actually do? the AI settings that change how your model behaves

Those sliders in your local AI app aren't decoration. Temperature and top-p control how creative or focused the model is — and getting them right… Aliteq Lena Fischer · AI & Local Compute Editor

Temperature = randomness/creativity. Low (0.1-0.3) = focused & factual; high (0.8-1.2) = creative & varied.

Low temp for: coding, math, data extraction, factual Q&A — you want the right answer, consistently.

High temp for: creative writing, brainstorming, roleplay — you want variety and surprise.

Top-p (nucleus): 0.9 is a good default — considers the most probable words up to 90% cumulative probability.

Top-k: limits choices to the top K words — another way to control randomness (40 is common).

Start with the defaults and adjust temperature first; it has the biggest effect.

Read the full story

Aliteq what do 'temperature' and 'top-p' actually do? the AI settings that change how your model behaves Read the full story on Aliteq