PLTAT/alpaca_bicolano

🇵🇭 Bicolano Alpaca Dataset

Dataset Summary

This dataset is a Bicolano translation of the original Alpaca instruction-following dataset. It is designed to support research and development of instruction-tuned language models for low-resource Philippine languages, particularly Bicolano.

The dataset retains the original Alpaca structure while providing high-quality translations of instructions, inputs, and outputs.

Dataset Structure

Data Instances

Each example follows this JSON format:

{
  "instruction": "Bicolano instruction text",
  "input": "Optional context in Bicolano",
  "output": "Expected response in Bicolano"
}

Data Fields

• instruction: The task or question in Bicolano.
• input: Additional context (may be empty).
• output: The correct expected response in Bicolano.

Data Splits

Split	Description
train	Main dataset for training
validation	Optional validation set (if provided)

Dataset Creation

Source Data

Based on the original Alpaca dataset, which was generated using instruction-following data derived from OpenAI models.

Translation Process

Translated from English to Bicolano using:

• Machine translation + human post-editing (or specify your actual method)
• Native speaker validation (if applicable)

Use Cases

This dataset can be used for:

• Instruction tuning of LLMs in Bicolano
• Multilingual NLP research
• Low-resource language modeling
• Chatbot and assistant development for Bicolano speakers

Limitations

• May contain translation artifacts or unnatural phrasing.
• Cultural nuances might not always be preserved.
• Not all instructions may perfectly align with Bicolano linguistic norms.
• Quality depends on the exact translation method used.

Ethical Considerations

Ensure responsible use when deploying models trained on this dataset. Be mindful of:

• Bias inherited from the original Alpaca dataset.
• Potential mistranslations or harmful outputs.
• Not intended for high-stakes applications without further validation.

Licensing

The original Alpaca dataset license applies. License: CC BY-NC 4.0 (Note: Datasets generated from OpenAI models are generally restricted from commercial use competing with OpenAI).

Citation

If you use this dataset, please cite:

@dataset{bicolano_alpaca,
  title = {Bicolano Alpaca Dataset},
  author = {Wely Jesch Sabalilag},
  year = {2026},
  note = {Translated version of the Alpaca dataset}
}

Acknowledgements

• OriginalAlpaca dataset creators (Stanford CRFM).
• Contributors and translators for Bicolano.

Contact

For questions or contributions:

• Name: Wely Jesch Sabalilag
• Email:welyjesch@gmail.com
• GitHub:github.com/welyjesch