Efficient-Large-Model/LongLive2.0-Toy-Dataset

该数据集是LongLive2.0发布代码的玩具格式检查数据集，旨在帮助用户在准备更大数据集之前验证AR扩散训练、DMD蒸馏和提示格式。数据集包含两个独立的训练文件夹：ar_training/用于AR扩散训练的配对视频/字幕数据，以及dmd_distillation/用于第二阶段DMD蒸馏的仅提示数据。它支持多种数据布局和配置选项，例如JSON提示目录或纯文本提示文件，以方便用户测试和调试代码。

This is a toy-format validation dataset accompanying the released code of LongLive2.0, intended to help users verify AR diffusion training, DMD distillation and prompt formatting before preparing larger-scale datasets. This dataset contains two separate training folders: `ar_training/` for paired video and subtitle data used in AR diffusion training, and `dmd_distillation/` for prompt-only data for the second-stage DMD distillation. It supports multiple data layouts and configuration options, such as JSON prompt directories or plain-text prompt files, to facilitate users' testing and debugging of the code.