rajtripathi/5M-Songs-Lyrics
Dataset Summary This dataset contains 50 million rows of song lyrics sourced from a public Kaggle dataset. It has been preprocessed into an instruction–label format suitable for training or fine-tuning generative language models, particularly for music lyric generation tasks. Each row is designed to guide a model to generate song verses in the style of a specific artist and genre, with corresponding real lyric snippets as ground truth. Supported Tasks and… See the full description on the dataset page: https://huggingface.co/datasets/rajtripathi/5M-Songs-Lyrics.
Dataset Summary
- This dataset contains 50 million rows of song lyrics sourced from a public Kaggle dataset. It has been preprocessed into an instruction–label format suitable for training or fine-tuning generative language models, particularly for music lyric generation tasks.
- Each row is designed to guide a model to generate song verses in the style of a specific artist and genre, with corresponding real lyric snippets as ground truth.
Supported Tasks and Benchmarks
- Text Generation: Fine-tuning LLMs to generate song verses given an artist and genre.
- Style Transfer: Capturing an artist’s lyrical style and applying it to a genre.
- Music AI Research: Studying stylistic patterns across artists and genres.
Languages
English (all lyrics are in lowercase after preprocessing)
Dataset Structure
Data Instances
Example row:
- Instruction Label Generate a song verse in the style of cam'ron in the genre of rap. [chorus: opera steve & cam'ron] killa cam, killa cam, killa cam ...
Data Fields
- Instruction (string): An instruction in the form:
"Generate a song verse in the style of {artistname} in the genre of {genrename}."
- Label (string): A snippet (up to 20,000 characters) from the song’s lyrics corresponding to the artist and genre.
Dataset Creation
Source Data
Original source: Public dataset from Kaggle containing 50 million song lyrics.
Processing steps:
- Dropped rows containing missing values.
- Converted all text to lowercase.
- Reformatted into two columns (Instruction, Label).
- Limited lyric text to a maximum of 20,000 characters for efficiency.
Uses
Direct Use
- Fine-tuning large language models (LLMs) for lyric generation.
- Creating artist–genre–lyric style datasets for music-related NLP tasks.
Out-of-Scope Use
- Any use that violates copyright laws in your jurisdiction.
- Commercial redistribution of original lyrics without rights.
Dataset Size
Rows: 50,000,000+
Columns: 2
File format: CSV
Licensing Information
The dataset contains copyrighted song lyrics. Use is intended for research and educational purposes only. Ensure you comply with relevant copyright laws when using the dataset.
Citation
If you use this dataset, please cite the Kaggle source:
Original dataset retrieved from Kaggle. This processed version reformats it for instruction–label learning tasks.
Acknowledgments
Thanks to the original Kaggle contributors for compiling the raw dataset. Enjoy working with it, and may it inspire creative AI-generated music! 🎵
