cyrille-elie/CHSA-Triage-Medic-Full-Dataset

该数据集是在AI工程师项目（CHSA项目）框架下构建的，旨在训练一个能够进行紧急分诊和提供临床推理的智能医疗助手。数据集分为三个不同的子集，对应于不同的训练阶段（监督微调和对齐）。

1. **sft_medical_dataset**：包含一般医学知识和临床案例（法语/英语），平衡为50%法语和50%英语，已通过Microsoft Presidio进行匿名化处理。
2. **sft_expert_dataset**：包含高质量的合成数据，专门用于紧急分诊教学，使用医学同义词数据增强技术以避免过拟合。
3. **dpo_dataset**：用于直接偏好优化（DPO）阶段，使模型学会偏好详细、结构化和临床准确的回答。

This dataset was created as part of an **AI Engineer** project (CHSA Project). It is designed to train an Intelligent Medical Assistant capable of performing **emergency triage** and providing clinical reasoning.

The dataset is divided into **3 distinct subsets** corresponding to different training phases (Supervised Fine-Tuning and Alignment).

1. **`sft_medical_dataset`**: General medical knowledge and clinical cases (FR/EN), balanced at 50% French and 50% English, anonymized via Microsoft Presidio.
2. **`sft_expert_dataset`**: Synthetic data specialized in emergency triage (FR), using medical synonym data augmentation techniques to avoid overfitting.
3. **`dpo_dataset`**: Preference pairs for alignment (Direct Preference Optimization), favoring detailed, structured, and clinically accurate responses.