Emova-ollm/emova-alignment-7m

EMOVA-Alignment-7M是一个用于全模态预训练的全面数据集，包括视觉语言和语音语言的对齐。该数据集使用了开源的图像文本预训练数据集、OCR数据集以及2000小时的自 curated ASR和TTS数据。它适用于多模态预训练，并包含了一般数据、OCR数据、纯文本指令数据和语音数据四个子集。EMOVA-Alignment-7M is a comprehensive dataset for omni-modal pre-training, including alignment of vision-language and speech-language. The dataset is constructed using open-sourced image-text pre-training datasets, OCR datasets, and 2,000 hours of self-curated ASR and TTS data. It is suitable for multi-modal pre-training and includes four subsets: general data, OCR data, pure textual instruction data, and speech data.