poemas.py

"""
### Spanish Poetry Dataset ###
Collection of Spanish poems retrieved by Andrea Morales and Miguel López from the website www.poemas-del-alma.com
Corpus adapted for Causal Language Modeling (CLM) to train GPT-like models. The author and title of each poem has been removed.
Note that, depending on your tokenizer, you might want to replace the <BOS>/<SEP>/<EOS> tokens by <|endoftext|> or something else.
Also note that the number of rows is slightly lower than the original dataset (andreamorgar/spanish_poetry) because a few incorrect examples have been filtered out.
"""

import datasets

_DESCRIPTION = "Collection of Spanish poems retrieved from www.poemas-del-alma.com"
_HOMEPAGE    = "https://www.kaggle.com/datasets/andreamorgar/spanish-poetry-dataset"
_AUTHORS     = "Andrea Morales and Miguel López"
_LICENSE     = "GNU Lesser General Public License"

class Poemas(datasets.GeneratorBasedBuilder):
    def _info(self):
        features = datasets.Features(
            {
                "text": datasets.Value("string"),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_AUTHORS,
        )

    def _split_generators(self, dl_manager):
        data_file = dl_manager.download_and_extract("poemas.txt")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": data_file,
                    "split": "train",
                },
            )
        ]

    def _generate_examples(self, filepath, split):
        to_replace = {"<BOS>": "", "<EOS>": "", "<SEP>": "\n"}
        with open(filepath, encoding="utf-8") as f:
            for key, poem in enumerate(f.readlines()):
                for old,new in to_replace.items():
                    poem = poem.replace(old, new)
                yield key, {"text": poem.strip()}