Temperature

Temperature is a configurable hyperparameter of generative language models that regulates the randomness and creativity of the model's output. At a low temperature value (e.g., 0.2), the model becomes deterministic, always choosing the next word with the highest probability, resulting in factual, precise, but repetitive responses. A higher value (e.g., 0.8–1.0) increases variety and 'risk-taking,' promoting creative, diverse text generation. Proper setting of this parameter is of fundamental importance depending on the use case, creating a balance between reliability and novelty.