--- license: mit --- # Bulgarian Jokes Dataset ## Overview The Bulgarian Jokes Dataset is a collection of Bulgarian-language jokes gathered and prepared for use in training and fine-tuning natural language processing (NLP) models. This dataset is designed to help researchers and developers build models capable of understanding and generating humorous content in Bulgarian. ## Dataset Structure The dataset is structured in a format suitable for NLP training and fine-tuning tasks, such as the Alpaca format. Each data entry contains fields that provide instructions, input, and expected output to facilitate supervised training of language models. ### Examples ```json { "instruction": "хуморески за Уиски", "input": "", "output": "Времената на сухия режим в САЩ.\r\nЖител на Ню Йорк отишъл в Тексас и попитал един местен жител:\r\n- Как я карате в тия тежки времена?\r\nОня отговорил:\r\n- В законите на нашия щат има една пролука. Имаш право да пиеш алкохол, само ако те ухапе гърмяща змия. Затова, за всеки случай всеки от нас носи в десния си джоб - бутилка уиски.\r\nНюйоркчанинът изтъкнал:\r\n- Но до такъв инцидент - може никога да не се стигне!\r\nТексасецът кимнал:\r\n- Точно така! Затова, всеки от нас носи в левия си джоб - една гърмяща змия!" }, { "instruction": "вицове за Знаете ли защо", "input": "", "output": "Знаете ли защо Нико Тупарев е кръстил новата си фиврма 24/7? Защото накой му е казал, че 24/7 е повече от 7/8 на Слави Трифонов." }, ``` ### Features Language: Bulgarian Categories: 599 categories covering a wide range of themes and topics Format: Suitable for training in Alpaca-like supervised formats ### Usage This dataset can be used for various NLP tasks, including: - Training and fine-tuning large language models (LLMs) to generate jokes in Bulgarian. - Evaluating the performance of language models in humor generation and understanding. - Creating NLP applications that require humorous content processing or generation. ### Sources The jokes in this dataset are collected from various sources such as humor websites, user-submitted content, and other public domain materials. Each entry was curated to ensure quality and diversity.