Alpaca-clean-archive
收藏资源简介:
这是斯坦福发布的原始羊驼数据集的清理版本。在原始版本中已发现以下问题,并在此数据集中进行了修复: 1.幻觉: 原始数据集中的许多指令都有引用互联网上数据的指令,这导致GPT3产生幻觉。 2.合并指令: 由于某些原因,在原始数据集中有许多指令被合并在一起。 3.空输出: 原始数据集中的某些条目具有空输出。 4.空代码示例: 原始数据集中的某些描述缺少代码示例,因此很难理解代码的预期行为。 5.生成图像的指令: 原始数据集中的一些描述包括生成图像的指令,这显然是不可能的。
This is a cleaned version of the original Alpaca dataset released by Stanford. The following issues identified in the original dataset have been fixed in this cleaned version: 1. Hallucinations: Many instructions in the original dataset referenced online data, which led GPT-3 to generate hallucinatory content. 2. Merged instructions: For unknown reasons, numerous instructions were combined together in the original dataset. 3. Empty outputs: Some entries in the original dataset had empty outputs. 4. Missing code examples: Certain descriptions in the original dataset lacked code examples, making it difficult to understand the intended behavior of the code. 5. Image generation instructions: Some descriptions in the original dataset included requests for image generation, which was clearly infeasible.




