llama2thedog/BIM-JEPA-FM-hf

BIM-JEPA-FM — IFCNetCore

BIM-JEPA-FM fine-tuned for 20-way IFCNetCore object classification. Provided for the inference demo.

• Paper: Toward generalizable foundation models for 3D BIM geometry using a joint embedding predictive architecture — Automation in Construction 191:107169 (2026)
• Code: jackswl/bim-jepa-fm
• Demo: BIM_JEPA_demo_2.ipynb

File	checkpoints/bim-jepa-fm-ifcnetcore.ckpt
Task	IFCNetCore classification, 20 classes
Backbone	12-layer transformer, 384 dim, 6 heads
Head	MLP, 256 dim, mean+max pooling
Fine-tuning	350 epochs from the BIM-JEPA-FM pre-trained encoder

Usage

from huggingface_hub import hf_hub_download
hf_hub_download(
    "llama2thedog/BIM-JEPA-FM-hf",
    "checkpoints/bim-jepa-fm-ifcnetcore.ckpt",
    local_dir="/content/",
)

The demo notebook loads this checkpoint through BimJepaClassification; the two model families are architecturally identical, so only the checkpoint path differs.

Citation

@article{shi2026toward,
  title={Toward generalizable foundation models for 3D BIM geometry using a joint embedding predictive architecture},
  author={Shi, Jack Wei Lun and Solihin, Wawan and Weng, Yufeng and Liang, Houhao and Zhao, Yimin and Poh, Leong Hien and Yeoh, Justin K.W.},
  journal={Automation in Construction},
  volume={191},
  pages={107169},
  year={2026},
  publisher={Elsevier}
}