benchang1110/

ShareGPT4V-zhtw

	ShareGPT4V-zhtw Dataset Card





	
	Dataset details

Dataset type: ShareGPT4V-zhtw is a Traditional Chinese version of Lin-Chen/ShareGPT4V translated by yentinglin/Llama-3-Taiwan-8B-Instruct. We took caption from Lin-Chen/ShareGPT4V, and discard captions of ocr-related images. We also remove the SFT captions, since this dataset is for feature-alignment only.

source number of captions

coco 168312

sam 590479

llava 588122

wikiart 500

share_textvqa 500… See the full description on the dataset page: https://huggingface.co/datasets/benchang1110/ShareGPT4V-zhtw.