Implementation:Explodinggradients Ragas BleuScore Metric

Field	Value
source	Repo
domains	Metrics, NLP
last_updated	2026-02-10

Overview

BleuScore computes the BLEU (Bilingual Evaluation Understudy) n-gram precision score between a reference and a generated response using the sacrebleu library.

Description

The BleuScore class evaluates text generation quality by computing BLEU score, which measures n-gram overlap between a reference and response. The implementation splits both reference and response into sentences (by ". " delimiter), then delegates to sacrebleu.corpus_bleu. The score is normalized to a 0-1 range by dividing by 100. This metric does not require an LLM and inherits only from SingleTurnMetric.

Key attributes:

• kwargs -- Additional keyword arguments passed to the underlying corpus_bleu function.

Dependency: Requires the sacrebleu package (pip install sacrebleu).

Usage

The metric requires reference and response columns.

Code Reference

Property	Value
Source Location	src/ragas/metrics/_bleu_score.py L11-49
Class Signature	class BleuScore(SingleTurnMetric)
Import	from ragas.metrics import BleuScore

I/O Contract

Inputs

Parameter	Type	Required	Description
reference	str	Yes	The ground truth reference text
response	str	Yes	The generated response to evaluate

Outputs

Output	Type	Description
score	float	BLEU score normalized to 0.0-1.0 range

Usage Examples

from ragas.metrics import BleuScore
from ragas.dataset_schema import SingleTurnSample

metric = BleuScore()

sample = SingleTurnSample(
    reference="The cat sat on the mat.",
    response="The cat is sitting on the mat."
)
# score = await metric.single_turn_ascore(sample)

Related Pages

• Explodinggradients_Ragas_RougeScore_Metric -- ROUGE-based n-gram overlap metric
• Explodinggradients_Ragas_ChrfScore_Metric -- Character n-gram F-score metric
• Explodinggradients_Ragas_StringMetrics_Module -- Non-LLM string comparison metrics