Fundamentals

Completion

A completion is the output text that an LLM generates in response to your prompt. The model predicts text token-by-token based on probability distributions learned during training. Completions can be deterministic or varied depending on parameters like temperature. The quality and relevance of completions depend on prompt clarity, model capabilities, and inference parameters. In practice, completions are often post-processed, validated, or regenerated until they meet your requirements.