ReactiveAI/ConversationalRetrieval-SMAT

该数据集是一个用于监督式记忆感知训练（SMAT）的对话检索数据集，包含多个配置，基于CoQA、NarrativeQA、Passkey Multitopic和QuAC等现有问答数据集构建。每个配置提供交互数据，特征包括查询（query）和答案（answer），部分配置还包含思考（think）字段，以支持模型在对话中进行推理和记忆。数据集旨在训练AI模型处理复杂的对话式问答任务，增强其上下文理解和多轮对话能力。具体配置涵盖指导式（instruct）、混合式（mix）和推理式（reasoning）变体，适用于不同的训练场景。

This dataset is a conversational retrieval dataset for Supervised Memory-Aware Training (SMAT), consisting of multiple configurations based on existing question-answering datasets such as CoQA, NarrativeQA, Passkey Multitopic, and QuAC. Each configuration provides interaction data with features including query and answer, and some configurations also include a think field to support model reasoning and memory in dialogues. The dataset is designed to train AI models for complex conversational question-answering tasks, enhancing their context understanding and multi-turn dialogue capabilities. Specific configurations cover instruct, mix, and reasoning variants, suitable for different training scenarios.