README.md

metadata

license: cc-by-nc-4.0
task_categories:
  - text-generation
  - question-answering
language:
  - es
tags:
  - instruction tuning
size_categories:
  - 10K<n<100K

Data Description

This HF data repository contains the Spanish Alpaca dataset used in our study of monolingual versus multilingual instruction tuning.

• GitHub
• Paper

Creation

• Machine-translated from yahma/alpaca-cleaned into Spanish.

Usage

• This data is intended to be used for Spanish instruction tuning.
• The dataset has roughly 52K instances in the JSON format.
• Each instance has an instruction, an output, and an optional input. An example is shown below:

{
"instruction": "Cuáles son los tres colores primarios?",
"input": "",
"output": "Los tres colores primarios son rojo, azul y amarillo. Estos colores se llaman primarios porque no pueden ser creados mezclando otros colores y todos los otros colores se pueden hacer combinando en varias proporciones. En el sistema de color aditivo, utilizado para la luz, los colores primarios son rojo, verde y azul (RGB)."
}

Known issues

• The machine translation process might have corrupted data containing code, cross-lingual tasks, grammatical error correction tasks, etc.

Citation

@inproceedings{chen-etal-2024-monolingual,
  title="Monolingual or multilingual instruction tuning: Which makes a better {Alpaca}",
  author="Pinzhen Chen and Shaoxiong Ji and Nikolay Bogoychev and Andrey Kutuzov and Barry Haddow and Kenneth Heafield",
  year="2024",
  booktitle = "Findings of the Association for Computational Linguistics: EACL 2024",
}