Skip to main content

luml.experiments.evaluation.scorers.builtin.prompt_alignment

PromptAlignment Objects​

class PromptAlignment(LLMJudgeScorer)

Scores whether a response follows the given instructions.

Unsupervised scorer — no expected output required. Considers format, length, style, and content constraints. Returns a float 0.0 (instructions ignored) to 1.0 (all instructions followed).

Default input keys: "instructions".

from luml.experiments.evaluation.scorers.builtin import PromptAlignment
scorer = PromptAlignment()
scorer = PromptAlignment(input_key="system_prompt")