wyu1/Leopard-Instruct

Leopard-Instruct是一个大规模的指令调优数据集，包含925K个实例，其中739K专门设计用于文本丰富的多图像场景。该数据集已用于训练Leopard-LLaVA和Leopard-Idefics2模型。

Leopard-Instruct is a large instruction-tuning dataset, comprising 925K instances, with 739K specifically designed for text-rich, multiimage scenarios. Its been used to train Leopard-LLaVA and Leopard-Idefics2 models. The dataset includes multiple subsets, each corresponding to different data file paths, supporting applications in multimodal, instruction-following, multi-image, language model, and vision-language model.