miso-choi/Allegator-train
收藏Hugging Face2024-10-19 更新2024-12-14 收录
下载链接：
https://hf-mirror.com/datasets/miso-choi/Allegator-train
下载链接
链接失效反馈官方服务：
资源简介：
该数据集用于Allegator微调的训练，由LLaVA-Instruct-150k和Flickr30k两个子集组成，分别包含99,883和31,783个样本。具体来说，LLaVA-Instruct-150k包含158k个语言-图像指令跟随样本，包括58k个对话、23k个描述和77k个复杂推理。为了增加视觉和文本的多样性，LLaVA-Instruct-150k被Flickr30k增强，后者具有相对较短的标题。
The dataset is used for training the finetuning of Allegator, consisting of subsets from LLaVA-Instruct-150k and Flickr30k, with 99,883 and 31,783 samples from each, respectively. In detail, LLaVA-Instruct-150K contains 158k language-image instruction following samples, including 58k conversations, 23k descriptions, and 77k complex reasoning. To enrich visual and textual diversity, LLaVA-Instruct-150K is augmented with Flickr30K, which has relatively short captions.
提供机构：
miso-choi
