Elvolvis/rlhf-dolly-conversations

该数据集包含4000多条高质量的指令-响应对和对话轨迹，这些数据来源于两个来源：Databricks Dolly 15k和HuggingFaceH4/UltraChat 200k。每条记录都经过格式化，用于微调语言模型在指令跟随和对话任务上的表现。数据集结构包括format_type（来源标识）、source_dataset（原始来源）、agentic_sequences（人类与助手之间的对话轮次）、category（任务类型）和context（背景信息，如果有的话）。

This dataset contains 4,000+ high-quality instruction-response pairs and conversational trajectories extracted from two sources: Databricks Dolly 15k and HuggingFaceH4/UltraChat 200k. Each record is formatted for fine-tuning language models on instruction-following and conversational tasks. The dataset structure includes format_type (source identifier), source_dataset (original source attribution), agentic_sequences (conversation turns between human and assistant), category (task type), and context (background information where available).