Munyarwanda-AI-Dataset v0.2

5,834 cleaned Kinyarwanda instruction examples (train: 5,542 / test: 292, split with test_size=0.05, seed=42).

Usage

from datasets import load_dataset
ds = load_dataset("arcange9/Munyarwanda-AI-Dataset", "v0.2")

Columns

Column	Description
system	Kinyarwanda system prompt (fixed)
instruction	"Andika igisubizo mu Kinyarwanda:"
input	empty string
output	Kinyarwanda text, max 800 chars

System Prompt

Uri Munyarwanda AI, umufasha w’ubwenge buhangano wibanda ku Kinyarwanda. Igisubizo cyawe kigomba kuba mu Kinyarwanda gisobanutse kandi gisanzwe. Ntukoreshe Icyongereza nk’ururimi rw’igisubizo keretse umukoresha agusabye mu buryo bweruye gusubiza mu Cyongereza cyangwa urundi rurimi.

Sources (5834 total)

• map-boy/kinyarwanda-dataset-v2 — 2,000 examples (general Kinyarwanda)
• DigitalUmuganda/Monolingual_health_dataset — 2,000 examples (Kinyarwanda health)
• KoseiUemura/KinNewsClassification — 1,834 examples (Kinyarwanda news)

Cleaning pipeline

• Unicode normalization (ftfy + NFC), whitespace collapse
• Kinyarwanda language detection (>60% Kinyarwanda-word ratio, English removed)
• Minimum 10 words per example, output capped at 800 chars
• Exact-hash deduplication
• Deterministic sampling (seed 42)

v0.1 (legacy)

The original 226-row starter dataset (instruction/response/category/source) remains available as config v0.1.