Modelpublic

fassabilf/sea-clip-tiny-init

sourceHugging Facemitupdated 1d agoView on Hugging Face

0likes16downloads

Model Card

SEA-CLIP-Tiny — English initialisation (e0)

The English CLIP-KD student checkpoint used to initialise every SEA-CLIP-Tiny run in SEA-CLIP-Tiny (ACCV 2026). It is the starting point, not a trained SEA model: no Southeast Asian data has been seen at this point.

Usage

python

import open_clip

model, _, preprocess = open_clip.create_model_and_transforms('hf-hub:fassabilf/sea-clip-tiny-init')
tokenizer = open_clip.get_tokenizer('hf-hub:fassabilf/sea-clip-tiny-init')

Model


Architecture	ViT-T/16 vision tower + 12-layer / 384-wide text tower, embed dim 512
Tokenizer	CLIP BPE, vocab 49408, context length 77
Parameters	46.11M (5.62M vision + 40.49M text)
Training data	English CLIP-KD pretraining (no SEA data)
Teacher	—

Reproduce

Training and evaluation code: https://github.com/fassabilf/sea-clip-tiny. The exact training configuration of this checkpoint is in params.txt in this repo.

Citation

bibtex

@inproceedings{seacliptiny2026,
  title     = {SEA-CLIP-Tiny: Efficient Multilingual Text-Vision Embedding for Southeast Asian Languages},
  booktitle = {Asian Conference on Computer Vision (ACCV)},
  year      = {2026}
}