Skip to content

Dataset · arbml

Culturally adapted Arabic instructions

arbml/CIDAR

10,000 manually verified instruction-response pairs in Arabic for SFT.

What's inside

The dataset includes 10,000 text instructions and responses in Arabic. Approximately 9,109 examples were translated from the Alpagasus dataset using ChatGPT, and another 891 Arabic grammar tasks were taken from the Ask the teacher website. All 10,000 entries were verified by a team of about 12 reviewers.

What you can do with it

  • SFT fine-tuning of language models for instruction-following in Arabic
  • Improving Arabic text generation quality and grammar in LLMs
  • Training specialized LoRA adapters for Arabic-speaking assistants
Tasks
Text generation
Languages
Arabic
Size
10K–100K rows
Downloads this month
2K
Likes
57

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("arbml/CIDAR")