Few-shot Examples

An in-context learning technique in large model applications. By providing several input-output examples in the prompt that closely resemble the target task, it enables the model to infer the required patterns, formats, and domain rules without updating parameters, offering low deployment cost and rapid iteration.