Feature/chrf score

[kauabh]

CHRF Score Metric Using sacrebleu

Issue Link / Problem Description

This PR introduces a new ChrfScore metric based on sacrebleu.corpus_chrf for evaluating the similarity between a generated response and a reference.
CHRF is better suited for morphologically rich languages and provides a character-level F-score.

Changes Made

Added ChrfScore class to implement character F-score (CHRF) metric using sacrebleu.corpus_chrf.

Testing

How to Test

• Manual testing steps:
  1. Install sacrebleu if not already installed: pip install sacrebleu
  2. Import and instantiate ChrfScore
  3. Pass a SingleTurnSample object with reference and response
  4. Run the metric and verify output is a float between 0.0 and 1.0

References

• sacrebleu documentation:

Screenshots/Examples (if applicable)

from sacrebleu import corpus_chrf

hypotheses = ["The cat is on the mat."]
references = [["The cat is sitting on the mat."]]

score = corpus_chrf(hypotheses, references).score / 100
print(score)  # e.g., 0.67