neph1/Unity_Code_QnA

该数据集基于Unity示例项目，围绕源代码文件展开，旨在不特定于提交的源代码文件，而是关注其中的通用编码模式。由于数据集是合成生成的，其结果部分依赖于LLM（在此案例中为Mistral-Nemo）的处理方式。因此，有时会询问关于“这段代码”的问题。为了获得最佳效果，训练时可能需要将源文件包含在上下文中。作者还提到会扩展该数据集，但需要大量的清理和剔除不良生成。

Multi response dataset based on Unity example projects, focusing on general coding patterns within source files. Since its synthetically generated, the results depend on the LLMs handling. Its recommended to include the source file in the context during training for best results. The dataset is still being expanded, involving a lot of cleaning and discarding of bad generations.