Netsoft/oai-instruct

--- task_categories: - text-generation language: - en size_categories: - 10K<n<100K --- ## OAI Instruct ### Overview OAI Instruct is a dataset designed for the fine-tuning of large language models, that includes comprehensive data structured to facilitate advanced natural language processing tasks. This dataset originates from the "5G INSTRUCT Forge" pipeline, which processes 3GPP specifications to generate training and testing data. The primary aim is to enable LLMs to comprehend and operate based on the intricacies of these technical specifications. ### Dataset Composition - **Training Set**: Contains 87,719 entries, with columns including: - `instruction`: Descriptive task guidance - `task_type`: Categorization of the instruction - `input`: Inputs relevant to the tasks - `completion`: Expected model outputs - **Test Set**: Comprises 9,557 entries, featuring: - `prompt`: Basis for generating completions - `completion`: Standard outputs for performance assessment - `completion2`: Alternative outputs for comparative evaluation ### Evaluation Metrics Performance evaluation in the test set utilizes well-known metrics, including: - **BERTScore**: Measures the semantic similarity between generated texts and reference texts. - **SemScore**: Provides a score based on semantic accuracy relative to the task requirements. ### Total Dataset Size - **Approximate Size**: 100MB ### Usage The OAI Instruct dataset is a key resource for developing LLMs capable of understanding and interacting with complex technical standards. It serves as a proof of concept for the "5G INSTRUCT Forge" pipeline. ### Repository For more detailed insights into the dataset generation and the overall pipeline, visit the GitLab repository: [5G INSTRUCT Forge Repository](https://gitlab.eurecom.fr/Azzedde1/5g_instruct_forge) ### Instructions for Import and Use (Using Hugging Face) To effectively utilize OAI Instruct with Hugging Face, follow these instructions: - **Set Up Environment**: Ensure you have Python and the `datasets` library from Hugging Face installed. If not, you can install it using pip: ```bash pip install datasets ``` - **Import the Dataset**: Use the `datasets` library to load the dataset directly from Hugging Face: ```python from datasets import load_dataset # Load the dataset dataset = load_dataset('Netsoft/oai-instruct') ``` - **Explore the Dataset**: Begin by exploring the dataset to understand its structure and contents: ```python # Print the dataset structure print(dataset) # Access the training set train_set = dataset['train'] print(train_set.column_names) print(train_set[0]) # Display the first entry # Access the test set test_set = dataset['test'] print(test_set.column_names) print(test_set[0]) # Display the first entry ### Citation When citing our dataset in your research, please use the following citation: **BibTeX**: ```bibtex @article{said20245g, title={5G INSTRUCT Forge: An Advanced Data Engineering Pipeline for Making LLMs Learn 5G}, author={Said, Azzedine Idir Ait and Mekrache, Abdelkader and Boutiba, Karim and Ramantas, Kostas and Ksentini, Adlen and Rahmani, Moufida}, journal={IEEE Transactions on Cognitive Communications and Networking}, year={2024}, publisher={IEEE} } ```任务类别： - 文本生成 语言： - 英语 规模类别： - 10K<n<100K --- ## OAI Instruct ### 概述 OAI Instruct是专为大语言模型（Large Language Model, LLM）微调设计的数据集，包含结构化的全面数据，可支撑各类高级自然语言处理任务。本数据集源自"5G INSTRUCT Forge"处理流水线，通过解析3GPP技术规范生成训练与测试数据，核心目标是使大语言模型能够理解并基于这些复杂技术规范开展工作。 ### 数据集构成 - **训练集**：共包含87,719条数据，字段如下： - `instruction`：任务描述性指导 - `task_type`：指令分类标签 - `input`：任务相关输入数据 - `completion`：模型预期输出结果 - **测试集**：共包含9,557条数据，字段如下： - `prompt`：生成输出的基准依据 - `completion`：用于性能评估的标准输出 - `completion2`：用于对比评估的备选输出 ### 评估指标 测试集的性能评估采用行业通用知名指标，包括： - **BERTScore**：衡量生成文本与参考文本之间的语义相似度 - **SemScore**：基于任务需求的语义准确性给出对应评分 ### 数据集总规模 - **近似大小**：100MB ### 应用场景 OAI Instruct数据集是开发能够理解并交互复杂技术标准的大语言模型的核心资源，同时也是"5G INSTRUCT Forge"处理流水线的概念验证示例。 ### 代码仓库 如需深入了解数据集生成流程与整体流水线架构，请访问以下GitLab仓库： [5G INSTRUCT Forge 仓库](https://gitlab.eurecom.fr/Azzedde1/5g_instruct_forge) ### Hugging Face 导入与使用指南 若需通过Hugging Face高效使用OAI Instruct数据集，请遵循以下步骤： - **环境配置**： 确保已安装Python及Hugging Face的`datasets`库，若未安装可通过pip执行以下命令： bash pip install datasets - **导入数据集**： 使用`datasets`库直接从Hugging Face加载该数据集： python from datasets import load_dataset # 加载数据集 dataset = load_dataset('Netsoft/oai-instruct') - **探索数据集**： 先探索数据集以了解其结构与内容： python # 打印数据集结构 print(dataset) # 访问训练集 train_set = dataset['train'] print(train_set.column_names) print(train_set[0]) # 显示第一条数据 # 访问测试集 test_set = dataset['test'] print(test_set.column_names) print(test_set[0]) # 显示第一条数据 ### 引用方式 若在研究工作中引用本数据集，请使用以下BibTeX格式： **BibTeX**： bibtex @article{said20245g, title={5G INSTRUCT Forge: An Advanced Data Engineering Pipeline for Making LLMs Learn 5G}, author={Said, Azzedine Idir Ait and Mekrache, Abdelkader and Boutiba, Karim and Ramantas, Kostas and Ksentini, Adlen and Rahmani, Moufida}, journal={IEEE Transactions on Cognitive Communications and Networking}, year={2024}, publisher={IEEE} }