README.md

---
pipeline_tag: sentence-similarity
language: hat
license: cc-by-nc-4.0
tags:
  - trimmed
library_name: sentence-transformers
base_model: facebook/metaclip-2-worldwide-giant-378
base_model_relation: quantized
datasets:
  - lbourdois/fineweb-2-trimming
---

# metaclip-2-worldwide-giant-378-hat-16384
This model is a **31.20% smaller** version of [facebook/metaclip-2-worldwide-giant-378](https://huggingface.co/facebook/metaclip-2-worldwide-giant-378) optimized for **Haitian Creole** language via vocabulary size reduction using the [trimming](https://huggingface.co/blog/lbourdois/introduction-to-trimming) method.  
This trimmed model should perform similarly to the original model with only 16,384 tokens and a much smaller memory footprint. However, it may not perform well for other languages as tokens not commonly used in the selected languages were removed from the vocabulary.

## Model Statistics
| Metric | Original | Trimmed | Reduction |
|--------|----------|---------|-----------|
| **Vocabulary size** | 901,629 tokens | 16,384 tokens | **98.18%** |
| **Model size** | 3,631,197,057 params | 2,498,083,457 params | **31.20%** |

![image](https://raw.githubusercontent.com/lbourdois/blog/refs/heads/master/assets/images/Trimming/metaclip-2-worldwide-giant-378-16384.png)

## Mining Dataset Statistics
- **Number of texts used for mining**: 200,000 texts  
- **Dataset**: [lbourdois/fineweb-2-trimming](https://huggingface.co/datasets/lbourdois/fineweb-2-trimming)

## Usage

#### Transformers (zero-shot image classification)
```python
from transformers import pipeline

# load pipeline
image_classifier = pipeline(model="alphaedge-ai/metaclip-2-worldwide-giant-378-hat-16384", task="zero-shot-image-classification")

# load image and candidate labels
image = "http://images.cocodataset.org/val2017/000000039769.jpg"
candidate_labels = ["Potential label 1 in Haitian Creole", "Potential label 2 in Haitian Creole", "Potential label 3 in Haitian Creole", "Potential label 4 in Haitian Creole"]

# run inference
outputs = image_classifier(image, candidate_labels)
print(outputs)
```

#### Sentence-transformers (texts-images similarity)
```python
from sentence_transformers import SentenceTransformer

model = SentenceTransformer("alphaedge-ai/metaclip-2-worldwide-giant-378-hat-16384")

images = [
    "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/car.jpg",
    "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg",
    "https://huggingface.co/datasets/huggingface/cats-image/resolve/main/cats_image.jpeg"
]
texts = ["Text 1 in Haitian Creole", "Text 2 in Haitian Creole", "Text 3 in Haitian Creole", "Text 4 in Haitian Creole"]

image_embeddings = model.encode(images)
text_embeddings = model.encode(texts)
print(image_embeddings.shape, text_embeddings.shape)

similarities = model.similarity(image_embeddings, text_embeddings)
print(similarities)
```

## Citations

#### Meta CLIP 2
```
@misc{chuang2025metaclip2worldwide,
      title={Meta CLIP 2: A Worldwide Scaling Recipe}, 
      author={Yung-Sung Chuang and Yang Li and Dong Wang and Ching-Feng Yeh and Kehan Lyu and Ramya Raghavendra and James Glass and Lifei Huang and Jason Weston and Luke Zettlemoyer and Xinlei Chen and Zhuang Liu and Saining Xie and Wen-tau Yih and Shang-Wen Li and Hu Xu},
      year={2025},
      eprint={2507.22062},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.22062}, 
}
```

#### Trimming blog post
```
@misc{hf_blogpost_trimming,
      title={Introduction to Trimming}, 
      author={Loïck BOURDOIS and Tom AARSEN and Bram VANROY and Christopher AKIKI and Woojun JUNG and Manuel ROMERO and Prithiv SAKTHI},
      year={2026},
      url={https://huggingface.co/blog/lbourdois/introduction-to-trimming}, 
}
```