README.md

metadata

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 for more details.

How to use

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

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)