HSQ-TD：Fitness and Health Qigong Large Model Fine-Tuning Dataset

This study constructs the first instruction fine-tuning dataset (HSQ-TD) for the field of fitness and health-preserving Qigong, aiming to address the prominent pain points of general large language models in this domain, including professional expression deviation, logical discontinuity, and potential medical safety risks. Based on authoritative corpora such as official textbooks issued by the General Administration of Sport of China and professional academic literature, the dataset undergoes rigorous text purification and structured instruction distillation, ultimately yielding 57,843 high-quality question-and-answer pairs covering core dimensions including movement specifications, scientific principles, and safety taboos.In the dataset construction process, this study innovatively introduces the Hybrid Chain-of-Thought (Hybrid CoT) reasoning mechanism.

Combined with Domain Consistency Index (DCI) filtering, Logical Chain Index (LCI) evaluation, and a dynamic safety alignment strategy, the reliability of the dataset in terms of professional knowledge depth and medical ethical boundaries is comprehensively guaranteed. Systematic experimental evaluation shows that the Qwen2.5-7B model fine-tuned on HSQ-TD achieves a remarkable win rate of 68.8% compared with the base model in a generative double-blind evaluation.