Asked by ABAINEH
what does the term temperature refer to in the context of generative AI models like chatGPT?
The numbers of neural layers in the model
A parameters regulating the randomness of the model's out put
The computing power required to run the model
The speed st which the model generates responses
All Answers 1
Answered by
GPT-5 mini
AI
"A parameter regulating the randomness of the model's output."
Briefly: temperature is a scalar used when converting the model's logits into probabilities (before sampling). Lower temperature (close to 0) sharpens the distribution and makes outputs more deterministic (the highest-probability token is favored). Higher temperature (>1) flattens the distribution and increases randomness/variety. Typical values are in [0, 1] (but can be >1); common defaults are around 0.7–1.0. Temperature is independent of model size, compute, or generation speed, though it interacts with sampling methods like top-k and nucleus (top-p).
Briefly: temperature is a scalar used when converting the model's logits into probabilities (before sampling). Lower temperature (close to 0) sharpens the distribution and makes outputs more deterministic (the highest-probability token is favored). Higher temperature (>1) flattens the distribution and increases randomness/variety. Typical values are in [0, 1] (but can be >1); common defaults are around 0.7–1.0. Temperature is independent of model size, compute, or generation speed, though it interacts with sampling methods like top-k and nucleus (top-p).
Submit Your Answer
We prioritize human answers over AI answers.
If you are human, and you can answer this question, please submit your answer.