BaixuW/aviation-llm-instruction-dataset

航空LLM指令数据集是一个早期阶段的数据集，用于在航空决策场景中对大型语言模型进行指令调优（SFT）。该数据集主要用于研究实验和框架验证，目标是使模型能够为多飞机操作生成结构化决策先验。数据集结构遵循指令格式，包括指令、输入和输出。数据构建基于基本的航空领域知识、简化的场景描述和基于模板的生成。目前数据集仍处于早期阶段，存在覆盖范围有限、场景简化、模板设计可能带来的偏见等局限性，不适合实际部署。未来工作可能包括扩展场景多样性、提高航空情境的真实性、增强结构化输出设计和融入更多专家知识。

The Aviation LLM Instruction Dataset is an early-stage dataset for instruction tuning (SFT) of large language models in aviation decision-making scenarios. This dataset is mainly designed for research experiments and framework validation, with the goal of enabling models to generate structured decision priors for multi-aircraft operations. The dataset follows an instruction format, including instruction, input, and output. Data construction is based on basic aviation domain knowledge, simplified scenario descriptions, and template-based generation. Currently, the dataset is in its early stages with limitations such as limited coverage, simplified scenarios, potential bias from template design, and is not suitable for real-world deployment. Future work may include expanding scenario diversity, improving realism of aviation situations, enhancing structured output design, and incorporating more expert knowledge.