π Summarization PreferenceΒΆ
metrics.llm_as_judge.direct.criteria.summarization_preference
CriteriaWithOptions(
name="summarization_preference",
description="Does the response capture the summary in the best possible way?",
options=[
CriteriaOption(
name="Excellent",
description="The response includes details such as key figures, numbers, dates and details which are crucial for the entire understanding.",
),
CriteriaOption(
name="Good",
description="The response includes statements expressing emotions and acclamations.",
),
CriteriaOption(
name="Average",
description="The order of events in the response follows a suitable chronological order.",
),
CriteriaOption(
name="Poor",
description="The response includes minor and irrelevant details which add no value in a summary.",
),
],
option_map={
"Excellent": 1.0,
"Good": 0.75,
"Average": 0.5,
"Poor": 0.0,
},
)
[source]from unitxt.llm_as_judge_constants import CriteriaOption
Read more about catalog usage here.