aya_evaluation_suite.py

from pathlib import Path
from typing import Dict, List, Tuple

import datasets
import pandas as pd

from seacrowd.utils import schemas
from seacrowd.utils.configs import SEACrowdConfig
from seacrowd.utils.constants import Licenses, Tasks

_CITATION = """
@misc{singh2024aya,
      title={Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning},
      author={Shivalika Singh and Freddie Vargus and Daniel Dsouza and Börje F. Karlsson and
      Abinaya Mahendiran and Wei-Yin Ko and Herumb Shandilya and Jay Patel and Deividas
      Mataciunas and Laura OMahony and Mike Zhang and Ramith Hettiarachchi and Joseph
      Wilson and Marina Machado and Luisa Souza Moura and Dominik Krzemiński and Hakimeh
      Fadaei and Irem Ergün and Ifeoma Okoh and Aisha Alaagib and Oshan Mudannayake and
      Zaid Alyafeai and Vu Minh Chien and Sebastian Ruder and Surya Guthikonda and Emad A.
      Alghamdi and Sebastian Gehrmann and Niklas Muennighoff and Max Bartolo and Julia Kreutzer
      and Ahmet Üstün and Marzieh Fadaee and Sara Hooker},
      year={2024},
      eprint={2402.06619},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
"""

_DATASETNAME = "aya_evaluation_suite"

_DESCRIPTION = """
Aya Evaluation Suite contains a total of 26,750 open-ended conversation-style
prompts to evaluate multilingual open-ended generation quality.
"""

_HOMEPAGE = "https://huggingface.co/datasets/CohereForAI/aya_evaluation_suite"

_LANGUAGES = ["ceb", "tha", "mya", "zsm", "jav", "ind", "vie", "sun", "ace", "bjn", "khm", "lao", "min"]

_LICENSE = Licenses.APACHE_2_0.value

_LOCAL = False

_URLS = {
    _DATASETNAME: "https://huggingface.co/datasets/CohereForAI/aya_evaluation_suite/resolve/main/dolly_machine_translated/test-00000-of-00001.parquet?download=true",
}

_SUPPORTED_TASKS = [Tasks.INSTRUCTION_TUNING]

_SOURCE_VERSION = "1.0.0"

_SEACROWD_VERSION = "2024.06.20"


class AyaEvaluationSuiteDataset(datasets.GeneratorBasedBuilder):
    """
    Aya Evaluation Suite contains a total of 26,750 open-ended conversation-style
    prompts to evaluate multilingual open-ended generation quality.
    """

    SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
    SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)

    BUILDER_CONFIGS = [
        SEACrowdConfig(
            name=f"{_DATASETNAME}_{LANG}_source",
            version=datasets.Version(_SOURCE_VERSION),
            description=f"{_DATASETNAME} {LANG} source schema",
            schema="source",
            subset_id=f"{_DATASETNAME}_{LANG}",
        )
        for LANG in _LANGUAGES
    ] + [
        SEACrowdConfig(
            name=f"{_DATASETNAME}_{LANG}_seacrowd_t2t",
            version=datasets.Version(_SEACROWD_VERSION),
            description=f"{_DATASETNAME} {LANG} SEACrowd schema",
            schema="seacrowd_t2t",
            subset_id=f"{_DATASETNAME}_{LANG}",
        )
        for LANG in _LANGUAGES
    ]

    DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_ind_source"

    def _info(self) -> datasets.DatasetInfo:

        if self.config.schema == "source":
            features = datasets.Features(
                {
                    "id": datasets.Value("int64"),
                    "inputs": datasets.Value("string"),
                    "targets": datasets.Value("string"),
                    "language": datasets.Value("string"),
                    "script": datasets.Value("string"),
                    "source_id": datasets.Value("int64"),
                }
            )

        elif self.config.schema == "seacrowd_t2t":
            features = schemas.text2text_features

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
        """Returns SplitGenerators."""

        data_path = Path(dl_manager.download_and_extract(_URLS[_DATASETNAME]))

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": data_path,
                    "split": "train",
                },
            ),
        ]

    def _generate_examples(self, filepath: Path, split: str) -> Tuple[int, Dict]:
        """Yields examples as (key, example) tuples."""

        language = self.config.name.split("_")[3]

        df = pd.read_parquet(filepath)
        df = df[df["language"] == language]

        for index, row in df.iterrows():
            if self.config.schema == "source":
                example = row.to_dict()

            elif self.config.schema == "seacrowd_t2t":
                example = {
                    "id": str(index),
                    "text_1": row["inputs"],
                    "text_2": row["targets"],
                    "text_1_name": "inputs",
                    "text_2_name": "targets",
                }

            yield index, example