hao05/llava-cot-5k-reannotated
收藏Hugging Face2026-04-23 更新2026-04-26 收录
下载链接：
https://hf-mirror.com/datasets/hao05/llava-cot-5k-reannotated
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个包含5000行多模态数据的子集，用于Speculators的训练和测试。数据来源于LLaVA-CoT-100k_relabel.vl.min10.jsonl的前5000行，包含4889张图像。数据集布局包括train.jsonl和images/目录下的图像资源。每行数据保留了原始的对话式监督，并指向对应的图像。
This dataset packages a 5k multimodal subset for Speculators training and testing. It contains 5000 rows and 4889 images, sourced from the first 5k rows of LLaVA-CoT-100k_relabel.vl.min10.jsonl. The layout includes train.jsonl plus image assets under images/. Each row keeps the original conversation-style supervision and points to the corresponding image under `images/`.
提供机构：
hao05
