BaixuW/aviation-llm-instruction-dataset

language:

zh en license: apache-2.0 taskcategories: text-generation question-answering taskids: instruction-tuning Aviation LLM Instruction Dataset

Dataset Summary

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 currently in a very early stage. It is mainly designed for research experiments and framework validation.

The goal is to enable models to generate structured decision priors for multi-aircraft operations.

Intended Use

This dataset is intended for:

Instruction tuning of LLMs Structured reasoning experiments Research on combining LLM with decision-making systems Example use cases include:

Conflict avoidance reasoning Weather-aware decision support Multi-agent coordination Dataset Structure

Each sample follows an instruction format:

{ "instruction": "Analyze the scenario and generate decision priors.", "input": "Scenario description...", "output": { "weatherprior": [], "conflictprior": [], "efficiencyprior": [], "turnprior": [] } } Data Construction

The dataset is constructed based on:

Basic aviation domain knowledge Simplified scenario descriptions Template-based generation Pipeline includes:

Scenario abstraction Prompt design Structured output formatting The construction process is still evolving and not fully standardized.

Limitations

Early-stage dataset with limited coverage Simplified scenarios Potential bias from template design Not suitable for real-world deployment Future Work

Future improvements may include:

Expanding scenario diversity Improving realism of aviation situations Enhancing structured output design Incorporating more expert knowledge Citation

@dataset{aviationinstructiondataset, title={Aviation LLM Instruction Dataset}, author={BaixuW}, year={2026} } Author

BaixuW