Aya Dataset
204k prompt-completion pairs written and curated by ~3,000 native-speaking contributors from 119 countries via the open-science Aya project — the largest fully human-annotated multilingual instruction dataset. Covers 65 languages (71 including dialects and scripts): 28 high-resource, 12 mid-resource, and 31 low-resource. Includes annotator demographics for bias research.
from datasets import load_dataset
ds = load_dataset("CohereLabs/aya_dataset")Preview a sample row
{
"inputs": "Jakie są trzy największe miasta w Polsce?",
"targets": "Trzy największe miasta w Polsce to Warszawa, Kraków i Łódź.",
"language": "Polish",
"language_code": "pol",
"annotation_type": "original-annotations"
}Fine-tune with this dataset
Estimated VRAM to fine-tune with QLoRA (4-bit base model + LoRA adapters), using conservative defaults:
Frequently asked questions
Can I use Aya Dataset commercially?
Yes — Aya Dataset is released under Apache 2.0, a permissive license that allows commercial use, including training models you ship in a product. Check the dataset card for attribution requirements before release.
How much data does Aya Dataset contain, and do I need all of it?
Aya Dataset contains 204k Rows. You rarely need all of it: for style and format fine-tuning, a few hundred to a few thousand examples are enough — load a slice (e.g. split="train[:1000]") and scale up only if quality plateaus.
What is Aya Dataset best used for?
Instruction-tuning in non-English languages with real human-written data. It belongs to the Multilingual section of our dataset hub, where you'll find alternatives and complementary sets.