README.md

---
library_name: transformers
license: cc-by-nc-sa-4.0
tags: []
---

## Model Details

- Model: ReasonCLIP-L14-336-S0-Rea
- Base model: [openai/clip-vit-large-patch14-336](https://huggingface.co/openai/clip-vit-large-patch14-336)
- Architecture: CLIP ViT-L/14
- Image resolution: 336
- Training stage: Stage 0 - Reasoning
- Training data: Only reasoning caption-image pairs from [ReasonLite-42M](https://huggingface.co/datasets/RISys-Lab/ReasonCLIPLite-42M) and [ReasonPro-16M](https://huggingface.co/datasets/RISys-Lab/ReasonCLIPPro-16M)

## Method

![Method overview](https://raw.githubusercontent.com/RISys-Lab/ReasonCLIP/main/doc/method.png)

## Resources

- GitHub: [RISys-Lab/ReasonCLIP](https://github.com/RISys-Lab/ReasonCLIP)
- Paper: [arXiv:2606.26794](https://arxiv.org/abs/2606.26794)

## Usage

```python
from transformers import CLIPModel, CLIPProcessor

model_id = "fesvhtr/ReasonCLIP-L14-336-S0-Rea"
model = CLIPModel.from_pretrained(model_id)
processor = CLIPProcessor.from_pretrained(model_id)
```

For the full checkpoint list, see the [ReasonCLIP model card](https://github.com/RISys-Lab/ReasonCLIP/blob/main/doc/model_card.md).