跳到主要内容Skip to main content

温度Temperature

Temperature

  • 模型能力Model capabilities
  • 第二站 · 上手Station 2 · Hands On
解决什么问题What problem it solves

控制模型输出的随机性。调低更稳,调高更有创意。Controls how random the output is. Lower is steadier; higher is more creative.

这是什么What it is

采样时控制「每次是否选最高概率的词」这个随机程度的参数。温度为0 时输出最确定。

A sampling parameter controlling how often the highest-probability token is chosen. A temperature of 0 gives the most deterministic output.

为什么需要知道Why it matters

写代码、抽取信息要低温;起名字、写文案可以高温。这是你能直接调的少数参数之一。Low temperature for code and extraction; high temperature for naming and copywriting. It is one of the few parameters you can adjust directly.

出处Sources
  • OpenAI 温度参数说明OpenAI temperature parameter guide ↗
  • Gemini 模型参数(温度)Gemini model parameters (temperature) ↗