BrepARG Pretrained Weights

Inference-only pretrained weights for BrepARG (CVPR 2026).

Each checkpoint contains only model_state_dict (no optimizer or training metadata).

Files

File	Description
checkpoint/weights/abc_ar.pt	AR model for ABC
checkpoint/weights/abc_vqvae.pt	SE VQ-VAE for ABC (codebook=8192)
checkpoint/weights/deepcad_ar.pt	AR model for DeepCAD
checkpoint/weights/deepcad_vqvae.pt	SE VQ-VAE for DeepCAD (codebook=4096)

Download

pip install -U huggingface_hub

huggingface-cli download qingtiannihao/BrepARG \
  checkpoint/weights/abc_ar.pt \
  checkpoint/weights/abc_vqvae.pt \
  checkpoint/weights/deepcad_ar.pt \
  checkpoint/weights/deepcad_vqvae.pt \
  --local-dir . \
  --local-dir-use-symlinks False

This creates checkpoint/weights/*.pt under your project root.

Citation

@article{li2026autoregressive,
  title={AutoRegressive Generation with B-rep Holistic Token Sequence Representation},
  author={Li, Jiahao and Bai, Yunpeng and Dai, Yongkang and Guo, Hao and Gan, Hongping and Shi, Yilei},
  journal={arXiv preprint arXiv:2601.16771},
  year={2026}
}