Implementation:Explodinggradients Ragas StringMetrics Module

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

Overview

The StringMetrics module provides lightweight, non-LLM string comparison metrics: ExactMatch, StringPresence, and NonLLMStringSimilarity with configurable distance measures.

Description

This module contains three simple string comparison metric classes and a supporting enum:

• ExactMatch -- Returns 1.0 if reference equals response exactly, 0.0 otherwise.
• StringPresence -- Returns 1.0 if the reference string is present anywhere within the response, 0.0 otherwise.
• NonLLMStringSimilarity -- Computes normalized string similarity using configurable distance measures from the rapidfuzz library. Returns 1 - normalized_distance.
• DistanceMeasure (Enum) -- Supported distance measures: LEVENSHTEIN, HAMMING, JARO, JARO_WINKLER.

All metric classes inherit only from SingleTurnMetric and do not require an LLM or embeddings.

Dependencies: NonLLMStringSimilarity requires the rapidfuzz package (pip install rapidfuzz).

Usage

All metrics require reference and response columns.

Code Reference

Property	Value
Source Location	src/ragas/metrics/_string.py L11-100
Class Signatures	class ExactMatch(SingleTurnMetric), class StringPresence(SingleTurnMetric), class NonLLMStringSimilarity(SingleTurnMetric)
Import	from ragas.metrics import ExactMatch, StringPresence, NonLLMStringSimilarity

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 (ExactMatch)	float	1.0 if exact match, 0.0 otherwise
score (StringPresence)	float	1.0 if reference is substring of response, 0.0 otherwise
score (NonLLMStringSimilarity)	float	Normalized similarity score (0.0 to 1.0) based on the selected distance measure

Usage Examples

from ragas.metrics import ExactMatch, StringPresence, NonLLMStringSimilarity
from ragas.metrics._string import DistanceMeasure
from ragas.dataset_schema import SingleTurnSample

# Exact match
exact = ExactMatch()
sample = SingleTurnSample(reference="Paris", response="Paris")
# score = await exact.single_turn_ascore(sample)  # 1.0

# String presence
presence = StringPresence()
sample = SingleTurnSample(reference="Paris", response="The capital is Paris, France.")
# score = await presence.single_turn_ascore(sample)  # 1.0

# String similarity with Levenshtein distance
similarity = NonLLMStringSimilarity(distance_measure=DistanceMeasure.LEVENSHTEIN)
sample = SingleTurnSample(reference="Paris", response="Pris")
# score = await similarity.single_turn_ascore(sample)  # ~0.8

Related Pages

• Explodinggradients_Ragas_SemanticSimilarity_Metric -- Embedding-based semantic similarity
• Explodinggradients_Ragas_BleuScore_Metric -- BLEU n-gram precision metric
• Explodinggradients_Ragas_RougeScore_Metric -- ROUGE n-gram overlap metric
• Explodinggradients_Ragas_ChrfScore_Metric -- Character n-gram F-score metric