llama3.1-8b-short-stories
收藏Llama 3.1 8B Short Stories 数据集概述
数据集基本信息
- 标签:llama 3.1, short story, creative writing
- 任务类别:text-generation
- 语言:en
数据集内容
- 来源:由 Llama 3.1 8B Instruct Q4KM 模型生成的短篇故事。
- 用途:
- 作为创意写作实验的基准
- 探索模型的审查、偏见和内容倾向
- 挖掘写作中应避免的常见陈词滥调和AI生成短语
数据集创建方法
-
种子提示:从 agentlans/euclaise-WritingPromptsX 数据集的每个短篇故事中随机选择三个连续的句子。
-
生成模板:
Write a short story inspired by the following excerpt, though you are not required to follow it strictly. The story should be creative and original, exploring any theme or genre you choose. Aim for a narrative that is engaging and emotionally resonant, with a clear beginning, middle, and end. Use a descriptive and immersive writing style, and maintain a tone that feels thoughtful and reflective. The story should be 800–1,200 words in length and suitable for a general audience.
[EXCERPT]
数据集字段
- seed:用作提示的原始三句摘录。
- text:Llama 3.1 生成的短篇故事输出。
- word_count:生成故事的字数。
- valid:布尔值,指示输出是否为完整的短篇故事。
示例条目
json { "seed": "But Michael stood motionless. Jack was silent. They both stared at themselves through each other.", "text": "The old wooden pier creaked beneath their feet as they stood at the waters edge, [...]", "word_count": 577, "valid": true }
局限性
- 模型生成的故事通常短于提示中指定的800–1,200字目标。
- 作为合成数据,部分故事可能缺乏润色或连贯的叙事结构。
- 用户在使用此数据集时应遵守 Llama 3.1 的使用指南。




