Dataset · davanstrien
Synthetic haiku dataset for DPO training
davanstrien/haiku_dpo
A set of synthetic prompts and haiku poems for training language models to follow genre structure.
What's inside
The dataset contains synthetically generated user queries and haikus created to improve the technical skills of LLMs in writing such poems. The corpus volume is between 10,000 and 100,000 examples in text and tabular formats.
What you can do with it
- Language model alignment using DPO for generating poems with strict syllable structure
- SFT fine-tuning of models on creative writing tasks in a given format
- Tasks
- Text generation, Reinforcement learning
- Size
- 10K–100K rows
- Downloads this month
- 273
- Likes
- 54
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("davanstrien/haiku_dpo")