The Aya Dataset is a multilingual instruction fine-tuning dataset curated by an open-science community via Aya Annotation Platform from Cohere Labs. The dataset contains a total of 204k human-annotated prompt-completion pairs along with the demographics data of the annotators.
This dataset can be used to train, finetune, and evaluate multilingual LLMs.
Curated by: Contributors of Aya Open Science Intiative.
Language(s): 65 languages (71 including dialects &… See the full description on the dataset page:
https://huggingface.co/datasets/CohereLabs/aya_dataset.
