Euroswarms/CommandNet

COMMANDNET军事科学数据集（1900-1999年）是一个用于历史军事分析的合成指令微调示例数据集。该数据集专注于一致的战略声音和明确的因果分析及反事实分析，涵盖常规和非对称背景，时间范围限制在1900-1999年的历史窗口。数据格式为ShareGPT JSONL，包含10,000行数据，每条记录是一个JSON对象，具有ShareGPT风格的对话和元数据，如年份、年代、战争类型、学说家族和军事科学标签。数据集围绕多个学说家族组织，包括工业消耗战与堑壕渗透、渗透与分散突击群、纵深作战、闪电战与机动合成兵种、两栖作战序列、持久人民战争、人口中心反叛乱、机动战与决策周期压力、空地一体战以及威慑与升级管理。质量特征包括历史边界检查、明确的因果和反事实部分、重复抑制以及适用于聊天微调工作流的ShareGPT格式。

COMMANDNET Military Science Dataset (1900-1999) is a synthetic instruction-tuning dataset for historical and doctrinal military analysis. It focuses on a consistent strategic voice and explicit causal analysis and counterfactual analysis, covering both conventional and asymmetric contexts, constrained to the historical window 1900-1999. The format is ShareGPT JSONL, with 10,000 rows, each record being a JSON object with ShareGPT-style conversation and metadata including year, decade, warfare_type, doctrine_family, and military_science_tags. The dataset is organized around doctrine families such as Industrial Attrition and Trench Penetration, Infiltration and Decentralized Assault Groups, Deep Operation, Blitz and Mobile Combined Arms, Amphibious Operational Sequencing, Protracted Peoples War, Population-Centric Counterinsurgency, Maneuver Warfare and Decision-Cycle Pressure, AirLand Battle, and Deterrence and Escalation Management. Quality characteristics include historical bound checks, explicit causal and counterfactual sections, duplicate suppression, and ShareGPT formatting for chat fine-tuning workflows.