FLORES-200 Benchmark

FLORES-200 Benchmark

MH Specific

Marathi — Human-translated evaluation benchmark for machine translation covering 200+ languages including Marathi, with 3,001 sentences from diverse web articles

Benchmark Marathi machine translation quality against 200 languages using the FLORES-200 evaluation set.
HomepageHuggingFaceGitHub

Quick Start

from datasets import load_dataset
ds = load_dataset('facebook/flores', 'mar_Deva')
print(f'Dev: {len(ds["dev"])}, DevTest: {len(ds["devtest"])}')
for ex in list(ds['devtest'])[:3]:
    print(f'[{ex["id"]}] {ex["sentence"][:80]}...')
Modality
Text (parallel, Marathi)
Size
3,001 sentences, 200+ languages
License
Format
CSV/JSON
Language
mr
Update Frequency
static
Organization
Meta AI
Category
benchmarks-fairness-dialects-tools

Schema

FieldTypeDescription
sentencestringSentence in source/target language
idintSentence ID aligned across 200 languages

Build With This

Create a Marathi MT evaluation framework using FLORES-200 with automatic metrics and human evaluation
Develop a cross-lingual transfer study measuring how Marathi MT benefits from related language data
Build a translation difficulty analyzer identifying which FLORES-200 sentences are hardest to translate to/from Marathi

AI Use Cases

Machine translation evaluationcross-lingual transfer benchmarking
Last verified: 2026-03-07