NarrativAI/CakrawalaRP

CakrawalaRP是一个纯合成的数据集，包含具有详细角色设定和背景故事的角色之间的丰富角色扮演对话。该数据集用于训练Cakrawala模型，这是Llama 3.1模型的微调版本，优化了生成沉浸式角色互动的能力。数据集包含13,000个对话对，每个对话包括角色描述、详细的环境描述以及包含动作、表情和情感描述的来回对话。每个对话至少有12-13轮。数据集专门设计用于训练模型以保持角色声音的一致性，生成丰富的描述性文本，并保持长篇对话的叙事连贯性。

The CakrawalaRP dataset is a purely synthetic dataset containing rich roleplaying conversations between characters with detailed personas and backstories. It was used to train Cakrawala models, a fine-tuned variant of Llama 3.1 models optimized for generating immersive character interactions. The dataset contains 13,000 conversation pairs, each including character descriptions, detailed environmental descriptions, and rich dialogues with descriptions of actions, expressions, and emotions. The dataset is designed to train models to maintain consistent character voices, generate rich descriptive text, produce natural flowing dialogue, and maintain narrative coherence across long conversations.