Text2CAD

该数据集包含了17万个CadQuery注释，旨在训练模型，使其能够从自然语言描述生成CAD代码。此外，该数据集被用于微调六种开源的大型语言模型（LLMs），在生成可执行的CAD程序方面表现出一致的改进。规模上，数据集包含了17万个注释，任务聚焦于文本到CAD代码的生成。

This dataset contains 170,000 CadQuery annotations, designed to train models to generate CAD code from natural language descriptions. Moreover, this dataset has been utilized to fine-tune six open-source large language models (LLMs), with consistent improvements observed in their performance of generating executable CAD programs. In terms of its scale, the dataset includes 170,000 annotations, and the task focuses on text-to-CAD code generation.

搜集汇总

背景与挑战

背景概述

Text2CAD数据集是一个专注于从自然语言生成CadQuery-based 3D模型的项目，支持NeurIPS提交。它包括数据标注、模型训练、推理和评估流程，涵盖了六个开源LLM模型，并提供了详细的步骤说明和评估方法。