Temperature

Temperature is a hyperparameter that controls how random or deterministic a model's outputs are.

• **Scale (typically 0-2):**
• 0: Deterministic, always picks most likely token
• 0.5-0.7: Balanced creativity and consistency
• 1.0: Default, moderate randomness
• 1.5-2.0: High creativity, more unexpected outputs

When to Use Low Temperature (0-0.3): - Factual Q&A - Code generation - Data extraction - Consistent outputs needed

When to Use High Temperature (0.8-1.5): - Creative writing - Brainstorming - Story generation - Diverse ideas wanted

• **Related Parameter:**
• Top-p (nucleus sampling): Alternative way to control randomness