agentlans/prompt-quality

Prompt Quality Assessment数据集是一个用于评估提示（prompt）质量的数据集，旨在帮助提高大型语言模型（LLM）的性能。数据集包含10万个从agentlans/chatgpt数据集中选取的提示，这些提示由多个不同的LLM独立评估，评分标准基于提示的清晰度、具体性和完整性。评估结果通过主成分分析（PCA）和逻辑函数转换，生成了一个连续的质量分数（0到1之间）。数据集展示了模型评分之间的高相关性、提示质量分数的分布情况，并提供了示例提示及其对应的质量分数。该数据集可用于训练提示质量分类器，改进提示工程方法，并提升用户交互和模型训练的数据质量。

The Prompt Quality Assessment dataset is designed to evaluate the quality of prompts, which strongly affects the performance of large language models (LLMs). The dataset consists of 100,000 prompts selected from the agentlans/chatgpt dataset, each independently evaluated by multiple different LLMs using a standardized rating template that assesses clarity, specificity, and completeness. The evaluation results were aggregated and standardized using principal component analysis (PCA) and a logistic function to produce a continuous quality score ranging from 0 to 1. The dataset demonstrates high positive correlations between model ratings, the distribution of transformed quality scores, and includes example prompts with their corresponding quality scores. This dataset can be used to train prompt-quality classifiers, improve prompt engineering methods, and enhance the quality of datasets used in user interactions and model training.