Temperature

A parameter that controls the randomness of a language model's output — higher values produce more creative responses, lower values produce more predictable ones. Temperature is a scalar applied to the logit distribution before sampling in a language model. At temperature 0 the model always picks the highest-probability token (greedy decoding, fully deterministic). At temperature 1 the model samples from the raw probability distribution. Above 1, low-probability tokens become more likely, increasing diversity and creativity at the cost of coherence.

Typical creative tasks use temperature 0.7–1.0; factual or code-generation tasks use 0.0–0.3. Temperature interacts with top-p (nucleus sampling) — both parameters are usually tuned together to balance quality and variety.

Temperature is a foundational concept in Prompting Technique. A parameter that controls the randomness of a language model's output — higher values produce more creative responses, lower values produce more predictable ones.