Overview
Temperature adjusts the probability distribution of the next token.
Effects
- Low Temperature (e.g., 0.1): The model is more predictable and factual, always choosing the most likely next word.
- High Temperature (e.g., 0.8+): The model is more creative and diverse, but also more prone to hallucinations and errors.
Use Cases
Use low temperature for coding or factual Q&A; use high temperature for creative writing or brainstorming.