Quick definition
Temperature scaling adjusts the sharpness of token probabilities.
- Category: Generation
- Focus: output style and randomness
- Used in: Lower randomness for precise, repeatable answers.
What it means
Lower temperature makes outputs more deterministic. In generation workflows, temperature scaling often shapes output style and randomness.
How it works
Generation settings control how the model samples tokens. They trade off creativity, determinism, and safety.
Why it matters
Generation settings trade off creativity, determinism, and safety.
Common use cases
- Lower randomness for precise, repeatable answers.
- Higher randomness for brainstorming and creative tasks.
- Stopping rules to end output at the right time.
Example
Use 0.2 for precise answers.
Pitfalls and tips
High randomness can reduce accuracy while low randomness can be repetitive. Tune per task and evaluate results.
In BoltAI
In BoltAI, this appears in model settings that shape responses.
