Sag69/bosnian-dataset

Ultimate Bosnian Dataset

Overview

This is a comprehensive Bosnian language dataset designed for training large language models. It combines literary works, dictionary entries, and bilingual content to create a rich training resource for Bosnian language AI applications.

Note: This dataset has been cleaned to remove poor English translations, fix paragraph formatting issues, and ensure high-quality content for optimal model training.

Dataset Statistics

• Total Entries: 258,138 (cleaned)
• Size: ~89MB
• Languages: Bosnian (primary), English (bilingual entries)
• Format: JSONL (JSON Lines)

Dataset Components

1. Literary Works (55,628 entries)

• Derviš i smrt (The Dervish and Death) by Meša Selimović
• Tvrđava (The Fortress) by Meša Selimović
• Na Drini ćuprija (The Bridge on the Drina) by Ivo Andrić
• Prokleta avlija (The Damned Yard) by Ivo Andrić
• Bosnian folk tales and stories

2. Dictionary Entries (40,712 entries)

• Complete Bosnian dictionary with grammatical information
• Parts of speech, gender, cases, etymology
• Example sentences and usage contexts
• Structured for language learning and understanding

3. Bilingual Dictionary (162,008 entries)

• Bosnian-English translations
• English-Bosnian reverse translations
• Definitions and explanations
• Multiple instruction formats for various use cases

Use Cases

Language Generation

• Generate Bosnian text in literary styles
• Create content in the style of Meša Selimović or Ivo Andrić
• Produce natural Bosnian language responses

Translation

• Bosnian to English translation
• English to Bosnian translation
• Idiomatic expression translation
• Context-aware translation

Language Understanding

• Bosnian grammar and vocabulary learning
• Dictionary lookup and definition
• Part-of-speech identification
• Etymology and word origins

Advanced Applications

• Style transfer for Bosnian text
• Question answering in Bosnian
• Text classification and analysis
• Language model fine-tuning

Dataset Format

Each entry follows this structure:

{
  "instruction": "Task description",
  "input": "Input text or context",
  "output": "Expected output or response",
  "author": "Author name (for literary works)",
  "source": "Source text or reference",
  "language": "bs/en",
  "type": "literary/dictionary/bilingual",
  "word": "Dictionary word (if applicable)",
  "direction": "translation direction (if applicable)"
}

Training Recommendations

Model Requirements

• Base model: Llama 3.2 3B Instruct or similar
• Minimum 8GB RAM for training
• Apple Silicon MPS or CUDA GPU recommended

Training Parameters

• Learning rate: 1e-4
• Epochs: 3-5 (avoid overfitting)
• LoRA rank: 16-32
• Batch size: 4-8 (adjust based on memory)

Anti-Repetition Settings

• repetition_penalty: 1.2
• no_repeat_ngram_size: 3
• early_stopping: True

License

This dataset is released under the MIT License.

MIT License

Copyright (c) 2024 Sag69

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Citation

If you use this dataset in your research or applications, please cite:

@dataset{ultimate_bosnian_dataset,
  title={Ultimate Bosnian Dataset for LLM Training},
  author={Sag69},
  year={2024},
  license={MIT},
  url={https://huggingface.co/datasets/Sag69/bosnian-dataset}
}

Contributing

Contributions to improve the dataset are welcome! Please submit issues or pull requests for:

• Additional literary works
• Dictionary corrections or additions
• Translation improvements
• New use cases or examples

Acknowledgments

• Literary works by Meša Selimović, Ivo Andrić, and other Bosnian authors
• Bosnian dictionary sources and contributors
• Bilingual dictionary contributors
• The Bosnian language community