veld-base / README.md
kimsan0622's picture
Add multilingual to the language tag (#1)
db26b82
---
language:
- en
- ko
- multilingual
license: apache-2.0
tags:
- vision, language
- pretrained model
- image-to-text
eos_token: </s>
---
# veld base
Pretrained Vision Encoder Text Decoder Model in Korean and English. See [Github](https://github.com/AIRC-KETI/veld) for more details.
## How to use
```python
from transformers import AutoProcessor, AutoModel
processor = AutoProcessor.from_pretrained("KETI-AIR/veld-base", trust_remote_code=True)
model = AutoModel.from_pretrained("KETI-AIR/veld-base", trust_remote_code=True)
```
You can use AutoTokenizer and AutoFeatureExtractor instead AutoProcessor.
You don't need to pass `trust_remote_code=True` for AutoTokenizer and AutoFeatureExtractor
```python
from transformers import AutoFeatureExtractor, AutoTokenizer, AutoModel
feature_extractor = AutoFeatureExtractor.from_pretrained("KETI-AIR/veld-base")
tokenizer = AutoTokenizer.from_pretrained("KETI-AIR/veld-base")
model = AutoModel.from_pretrained("KETI-AIR/veld-base", trust_remote_code=True)
```