X-JEPA Model Checkpoints

This repository hosts the pretrained checkpoints for the paper:

Latent Prediction Needs Alignment: A Controlled Study of Joint-Embedding Predictive Vision-Language Learning
Mohammad Kohankhaki, Daniel Kusuma, Shirin Salehi, Carsten Kamp, Sigrid Brell-Cokcan, and Anke Schmeink — to appear at AACL-IJCNLP 2026

📦 Code: github.com/mohkoh19/x-jepa

---

Checkpoints

| File | Model | Size | |---|---|---|---| | clip.ckpt | CLIP baseline | 2.2 GB | | siglip.ckpt | SigLIP baseline | 2.2 GB | | xjepa_p.ckpt | X-JEPA [P] — prediction only | 3.0 GB | | xjepa_tc.ckpt | X-JEPA [TC] — target-contrastive | 3.1 GB | | xjepa_pa_lam003.ckpt | X-JEPA [P,A] λ=0.03 | 3.1 GB | | xjepa_pa_lam01.ckpt | X-JEPA [P,A] λ=0.1 (main model) | 3.1 GB | | xjepa_pa_lam03.ckpt | X-JEPA [P,A] λ=0.3 | 3.1 GB | | xjepa_pa_lam10.ckpt | X-JEPA [P,A] λ=1.0 | 3.1 GB |

Download

# Clone the code repo
git clone https://github.com/mohkoh/x-jepa.git
cd x-jepa

# Download all checkpoints
bash scripts/download_checkpoints.sh

Or download a single file:

wget https://huggingface.co/mohkoh/x-jepa/resolve/main/xjepa_pa_lam01.ckpt

Main Results (Paper Table 1)

Model	COCO ZS MR	Flickr ZS MR	SugarCrepe++	SVO Acc	VSR AUROC	NLVR2 Token
CLIP	67.89	79.87	71.68	84.43	63.75	54.93
SigLIP	67.67	80.32	69.79	84.30	62.77	55.00
X-JEPA [P]	0.10	0.21	37.06	50.36	48.52	53.05
X-JEPA [TC]	44.60	48.30	44.17	80.73	57.26	56.42
X-JEPA [P,A] λ=0.1	69.39	81.53	73.30	85.13	63.91	60.11

Citation

@inproceedings{kohankhaki2026latent,
  title     = {Latent Prediction Needs Alignment: A Controlled Study of Joint-Embedding Predictive Vision-Language Learning},
  author    = {Kohankhaki, Mohammad and Kusuma, Daniel and Salehi, Shirin and Kamp, Carsten and Brell-Cokcan, Sigrid and Schmeink, Anke},
  booktitle = {Proceedings of AACL-IJCNLP 2026},
  year      = {2026},
  publisher = {Association for Computational Linguistics},
}