PersonalLLM: A Benchmark for Personalizing LLMs

This dataset, presented in PersonalLLM: Tailoring LLMs to Individual Preferences, focuses on adapting LLMs to individual user preferences. It provides open-ended prompts paired with multiple high-quality responses, allowing for the evaluation of personalization algorithms. The dataset includes diverse user preferences simulated using pre-trained reward models, offering a robust testbed for research in this area.

The data is structured to handle continual data sparsity, a common challenge in personalized LLM applications. The dataset includes both training and evaluation sets.

For details on the dataset features, please refer to the metadata section above.

Getting Started

See the Github repository for detailed instructions on using this dataset and evaluating personalization algorithms. (Please replace "..." with the actual Github repository URL.)