clip-vit-large-patch14-ko
| Property | Value |
|---|---|
| Parameter Count | 428M parameters |
| Model Type | CLIP Vision-Language Model |
| Architecture | ViT-Large-Patch14 |
| License | MIT |
| Paper | Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation |
What is clip-vit-large-patch14-ko?
clip-vit-large-patch14-ko is a Korean-language adaptation of the CLIP (Contrastive Language-Image Pre-training) model, specifically designed for zero-shot image classification tasks. Developed by Bingsu, this model leverages knowledge distillation techniques to enable multilingual capabilities while maintaining the powerful vision-language understanding of the original CLIP architecture.
Implementation Details
The model is built on the ViT-Large architecture with 14x14 patch size, containing 428M parameters. It was trained using Korean-English parallel data from AIHUB, implementing the knowledge distillation methodology described in the original paper. The model supports both PyTorch and TensorFlow frameworks and is available in F32 and I64 tensor formats.
- Trained on comprehensive Korean-English parallel datasets from AIHUB
- Implements vision transformer architecture with 14x14 patch size
- Supports zero-shot classification capabilities
- Available in Safetensors format
Core Capabilities
- Zero-shot image classification with Korean text descriptions
- Multi-modal understanding between Korean text and images
- Flexible implementation with major deep learning frameworks
- Efficient inference with pre-trained weights
Frequently Asked Questions
Q: What makes this model unique?
This model is specifically optimized for Korean language understanding in vision-language tasks, making it one of the few CLIP models that can effectively process Korean text descriptions for image classification.
Q: What are the recommended use cases?
The model excels at zero-shot image classification tasks where Korean language descriptions are needed. It's particularly useful for applications requiring image understanding with Korean text queries, content moderation, and automated image categorization systems.
