TrustLLM: Trustworthiness in Large Language Models

• [12/01/2024] 🏄 The dataset, leaderboard, evaluation toolkit are released!

Table of Content

• About
• Dataset & Task
  • Dataset overview:
  • Task overview:
• Before Evaluation
  • Installation
  • Dataset Download
  • Generation
• Evaluation
• Leaderboard
• Contribution
• Citation
• License

We introduce TrustLLM, a comprehensive study of trustworthiness in LLMs, including principles for different dimensions of trustworthiness, established benchmark, evaluation, and analysis of trustworthiness for mainstream LLMs, and discussion of open challenges and future directions. Specifically, we first propose a set of principles for trustworthy LLMs that span eight different dimensions. Based on these principles, we further establish a benchmark across six dimensions including truthfulness, safety, fairness, robustness, privacy, and machine ethics. We then present a study evaluating 16 mainstream LLMs in TrustLLM, consisting of over 30 datasets. The document explains how to use the trustllm python package to help you assess the performance of your LLM in trustworthiness more quickly. For more details about TrustLLM, please refer to project website.

✓ the dataset is from prior work, and ✗ means the dataset is first proposed in our benchmark.

○ means evaluation through the automatic scripts (e.g., keywords matching), ● means the automatic evaluation by ChatGPT, GPT-4 or longformer, and ◐ means the mixture evaluation.

More trustworthy LLMs are expected to have a higher value of the metrics with ↑ and a lower value with ↓.

Installation via pip:

pip install trustllm

Installation via conda:

conda install -c conda-forge trustllm

Installation via Github:

git clone [email protected]:HowieHwong/TrustLLM.git

Creat a new environment:

conda create --name trustllm python=3.9

Install required packages:

cd trustllm_pkg
pip install .

Download TrustLLM dataset:

from trustllm.dataset_download import download_huggingface_dataset

download_huggingface_dataset(save_path='save_path')

The datasets are structured in JSON format, where each JSON file consists of a collection of dict(). Within each dict(), there is a key named prompt. Your should utilize the value of prompt key as the input for generation. After generation, you should store the output of LLMs as s new key named res within the same dictionary. Here is an example to generate answer from your LLM:

For each dataset, we have configured the temperature setting during model generation. Please refer to this page for the settings.

import json

filename = 'dataset_path.json'

# Load the data from the file
with open(filename, 'r') as file:
    data = json.load(file)

# Process each dictionary and add the 'res' key with the generated output
for element in data:
    element['res'] = generation(element['prompt'])  # Replace 'generation' with your function

# Write the modified data back to the file
with open(filename, 'w') as file:
    json.dump(data, file, indent=4)

See our docs for more details.

If you want to view the performance of all models or upload the performance of your LLM, please refer to this link.

We welcome your contributions, including but not limited to the following:

• New evaluation datasets
• Research on trustworthy issues
• Improvements to the toolkit

If you intend to make improvements to the toolkit, please fork the repository first, make the relevant modifications to the code, and finally initiate a pull request.

• Faster and simpler evaluation pipeline
• Dynamic dataset
• More fine-grained datasets
• Chinese output evaluation
• Downstream application evaluation

@misc{sun2024trustllm,
      title={TrustLLM: Trustworthiness in Large Language Models}, 
      author={Lichao Sun and Yue Huang and Haoran Wang and Siyuan Wu and Qihui Zhang and Chujie Gao and Yixin Huang and Wenhan Lyu and Yixuan Zhang and Xiner Li and Zhengliang Liu and Yixin Liu and Yijue Wang and Zhikun Zhang and Bhavya Kailkhura and Caiming Xiong and Chaowei Xiao and Chunyuan Li and Eric Xing and Furong Huang and Hao Liu and Heng Ji and Hongyi Wang and Huan Zhang and Huaxiu Yao and Manolis Kellis and Marinka Zitnik and Meng Jiang and Mohit Bansal and James Zou and Jian Pei and Jian Liu and Jianfeng Gao and Jiawei Han and Jieyu Zhao and Jiliang Tang and Jindong Wang and John Mitchell and Kai Shu and Kaidi Xu and Kai-Wei Chang and Lifang He and Lifu Huang and Michael Backes and Neil Zhenqiang Gong and Philip S. Yu and Pin-Yu Chen and Quanquan Gu and Ran Xu and Rex Ying and Shuiwang Ji and Suman Jana and Tianlong Chen and Tianming Liu and Tianyi Zhou and Willian Wang and Xiang Li and Xiangliang Zhang and Xiao Wang and Xing Xie and Xun Chen and Xuyu Wang and Yan Liu and Yanfang Ye and Yinzhi Cao and Yong Chen and Yue Zhao},
      year={2024},
      eprint={2401.05561},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

The code in this repository is open source under the MIT license.