luml.experiments.evaluation.scorers.builtin.summarization

Summarization Objects

class Summarization(LLMJudgeScorer)

Scores how well a summary captures its source text.

Unsupervised scorer — no expected output required. Evaluates accuracy, coverage, and absence of hallucination. Returns a float 0.0 (inaccurate / fabricated) to 1.0 (accurately captures all key information).

Default input keys: "text".

from luml.experiments.evaluation.scorers.builtin import Summarization
scorer = Summarization()
scorer = Summarization(input_key="document")