Chunjiang-Intelligence/OpenSCAD_3D_SFT
收藏资源简介:
该数据集是一个用于OpenSCAD监督微调(SFT)模型的合成生成SFT对话语料库。它由参数级突变和创造性建模任务两部分组成,比例约为8:2。参数级突变部分涉及对种子示例在维度、位置和基数轴上进行受控扰动,用于直接代码生成任务;创造性建模部分则包含几何复杂问题,通过显式的逐步链式思维(CoT)推理解决,用于推理增强生成任务。数据集的目的是平衡模型对OpenSCAD语法规律的学习和组合推理能力的发展,支持从中文或英文自然语言规范生成有效、可编译且参数化的OpenSCAD源代码。
This dataset is a synthetic generative SFT dialogue corpus for OpenSCAD supervised fine-tuning (SFT) models. It consists of two parts: parameter-level mutation tasks and creative modeling tasks, with a ratio of approximately 8:2. The parameter-level mutation section involves applying controlled perturbations to seed examples along dimensions, positions, and cardinal axes, which is designed for direct code generation tasks; the creative modeling section includes geometrically complex problems solved via explicit step-by-step chain-of-thought (CoT) reasoning, targeting reasoning-augmented generation tasks. The goal of this dataset is to balance the model's learning of OpenSCAD grammatical rules and the development of its combinatorial reasoning capabilities, while supporting the generation of valid, compilable, and parameterized OpenSCAD source code from natural language specifications in either Chinese or English.




