What is Temperature?
TL;DR
A parameter that controls the diversity and creativity of AI outputs. Lower values produce certainty, higher values produce creativity.
Temperature: Definition & Explanation
Temperature is a parameter that controls the randomness (diversity) of an LLM's text generation. It typically ranges from 0 to about 2. Lower values (closer to 0) cause the model to select the highest-probability tokens, producing consistent and reliable output. Higher values increase randomness, generating more creative and diverse outputs. Appropriate settings depend on the use case: low temperature (0-0.3) is best for factual answers and code generation, while high temperature (0.7-1.0) suits creative writing and brainstorming. At temperature 0, the output is nearly identical each time, making it useful for tasks requiring reproducibility. It is often used in conjunction with the top-p parameter.