Token Sampling

Definitions

1. The process of selecting output tokens from a probability distribution during generation rather than always taking the single highest-probability choice. Token sampling allows variety and creativity but can reduce determinism.

   In plain English: Choosing generated tokens by sampling from probabilities instead of always taking the top one.

   Example: "They disabled aggressive token sampling in the compliance workflow because the feature needed repeatable structured output."

Related Terms

• AI Generation
• AI Image
• Generative Search
• LLM Completion
• Non-Deterministic Output
• Stop Sequence
• Structured Generation
• Token Generation