GPU编程问答对

## 国科大 -《GPU架构与编程》特供微调数据集（仅加分题） ## 数据集基于GPU编程书籍、文档（cuTile-Triton-Tilelang等），通过GPT-5.2、Gemini-3-pro等LLM进行语料清洗问答 ## 不保证问答对的绝对正确性，但具有参考价值 ## 问答对格式：instruction对应system_promt，question对应input，answer对应output ## LORA参数参考：秩-64；缩放因子-128University of Chinese Academy of Sciences - Exclusive Fine-tuning Dataset for *GPU Architecture and Programming* (Only for Bonus Questions) This dataset is developed based on GPU programming books and documents including cuTile, Triton, Tilelang, etc., with its question-answer corpus cleaned and curated by LLMs such as GPT-5.2 and Gemini-3-pro. The absolute correctness of the question-answer pairs is not guaranteed, but they have reference value. The format of question-answer pairs is: instruction corresponds to system_prompt, question corresponds to input, and answer corresponds to output. Reference LoRA parameters: rank - 64; scaling factor - 128

搜集汇总

背景与挑战

背景概述

该数据集基于GPU编程书籍和文档，使用GPT-5.2等大型语言模型生成问答对，但准确性未作保证。数据格式与LoRA参数已提供，可作为参考资源。