Temperature: Definition

Temperature is a value between 0 and 1 (sometimes higher), and it adjusts how the model samples from its possible next words. (aka. deterministic vs stochastic)

🧠 How It Works

  • Low temperature (e.g., 0.0–0.3):

    • The output is more deterministic and focused.

    • The model picks the most likely next word almost every time.

    • Useful for factual, reliable, or repetitive tasks.

  • Medium temperature (e.g., 0.7):

    • The model becomes more creative and varied.

    • Still generally coherent and relevant.

  • High temperature (e.g., 1.0+):

    • The model produces more diverse and unpredictable outputs.

    • Can be creative but also less accurate or even nonsensical.