Temperature Parameter
Updated 2026-08-09
INTRODUCTION
English translation pending.
CORE DEFINITION
The temperature parameter scales the logits before the softmax, controlling how random a model's output distribution becomes. The core claim is that a high temperature flattens the distribution and produces varied, novel, occasionally incoherent output, while a low temperature sharpens it toward the most probable option and produces safe, repetitive results. The qualification is that the right setting is task-dependent: accuracy and exploration are traded off against each other, so the same value is not appropriate everywhere.
SCAFFOLDING EFFECT
Reduce cognitive load
- Creativity Dial: Deliberately adjust your own temperature, cooling for accuracy and heating for surprise. - Task Matching: Financial reporting calls for a low setting, while brainstorming calls for a high one. - Range Awareness: Many people stay mediocre by running cold constantly, and many burn out by running hot constantly.
Anchor fast decisions
In machine learning, temperature scales the logits before the softmax. A large value flattens the output distribution, making sampling more random and creative, while a small value concentrates probability on the most likely option, making output deterministic and conservative. In physics the same quantity measures the intensity of molecular thermal motion.
MINIMUM ACTION
In progress 0/1Practice this model in one real situation:
account_treeGenealogyexpand_more
menu_bookReferencesexpand_more
Source support: Explicit
- en.wikipedia.orghttps://en.wikipedia.org/wiki/Softmax_functionverified
PRIVATE NOTES · Only visible to you
SAVED Q&A
ENTRY Q&A · Private saving available
Ask with a clear boundary
thinkingmodels answers from published entry context only.
Your question is sent to thinkingmodels. The answer uses public entry context only.
RELATED MODELS