PersonalLLM

PersonalLLM is designed to facilitate research in Large Language Model (LLM) personalization. Unlike traditional unimodal preference-style datasets, PersonalLLM offers a diverse set of prompts and responses that reflect a wide range of user preferences. Additionally, we provide an evaluation set for various personalization algorithms.

Attached is a recommendation systems inspired method for learning across users for personalization.

Getting Started

Environment Setup

To set up the environment, you can use either poetry (preferred) or conda with the provided env.yml file or requirements.txt:

For poetry, run:

poetry install
poetry shell

For conda, run:

conda create --name personalllm python=3.10
conda env update --file env.yml --prune

Personalization Algorithm Evaluation

We'll love to see you contribute new personalization algorithms! They are designed to be as easily contributable as possible, refer to

• Evaluation Guide
• Zero Shot Algorithm
• K Shot ICL Algorithm

Dataset Generation

For detailed instructions on generating the dataset, please see the Dataset Generation Guide.

Dataset Samples

• PersonalLLM
• PersonalLLM-Eval

Paper

All plots were generated using the code in paper/visualize.ipynb file.

Contact:

Please email andrew.siah@columbia.edu for any help and open an issue for any bugs.