BlenderLLM: Training Large Language Models for Computer-Aided Design with Self-improvement

This paper introduces BlenderLLM, a large language model trained using a self-improvement approach specifically for computer-aided design (CAD) tasks. We developed a customized training dataset called BlendNet and introduced a comprehensive evaluation suite, CADBench. Results demonstrate that existing models exhibit significant limitations in generating accurate CAD scripts, while BlenderLLM achieves superior functionality and accuracy with minimal instruction-based fine-tuning and iterative self-improvement.

Download Paper

Recommended citation: Du, Y., Chen, S., Zan, W., Li, P., Wang, M., Song, D., Li, B., Hu, Y., & Wang, B. (2024). "BlenderLLM: Training Large Language Models for Computer-Aided Design with Self-improvement." arXiv preprint arXiv:2412.14203.
Download Paper