Temperature

Temperature changes how sharp or flat a distribution is before sampling.

Low temperature makes high-probability tokens even more dominant. High temperature spreads probability more widely across options.

Plain reading:

lower temperature -> safer, less varied
higher temperature -> more varied, more risk

Temperature does not add knowledge to the model. It changes how strongly the decoding process follows the model's preferred tokens.

Exercise

If temperature is lowered, should the distribution usually become sharper? Answer 1 for yes or 0 for no.