Conversations.py

import datasets
import os
from datasets import Dataset, DatasetDict
import gzip
import json


_CITATION = """\
@misc{Conversations,
  author       = {Ilya Koziev},
  title        = {Russian-Language Dialogues Dataset},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\\url{https://huggingface.co/datasets/inkoziev/Conversations}},
}
"""


_DESCRIPTION = """\
Russian-Language Dialogues Dataset
"""


class DatasetConfig(datasets.BuilderConfig):
    def __init__(self, **kwargs):
        super(DatasetConfig, self).__init__(**kwargs)



class Conversations(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIG_CLASS = DatasetConfig
    BUILDER_CONFIGS = [
        DatasetConfig(name="Conversations",
                      version=datasets.Version("27.02.2025"),
                      description=_DESCRIPTION,
                      ),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
               {
                   "domain": datasets.Value("string"),
                   "conversation": datasets.Value("string"),
               }
            ),
            supervised_keys=None,
            homepage="",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        download_url = "https://huggingface.co/datasets/inkoziev/Conversations/resolve/main/conversations.jsonl.gz"
        #download_url = "/home/inkoziev/github/Conversations/conversations.jsonl.gz"
        path = dl_manager.download(download_url)
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"datapath": path},
            )
        ]

    def _generate_examples(self, datapath):
        with gzip.open(datapath, "rt", encoding="utf-8") as f:
            for iline, line in enumerate(f):
                yield iline, json.loads(line)