RareConcepts/ZImage-Turbo-200k-multires-aspectbucketed

ZImage-Turbo WebDataset是一个使用DiffusionDB提示从ZImage-Turbo模型生成的图像数据集。数据集包含211,081个样本，分为216个分片，每个分片约1000个样本。生成过程使用了8个NVIDIA RTX 3090 GPU，耗时约2天，成本约为70美元。数据集提供了多种基础分辨率和纵横比的配置，每个分片包含JPEG-XL编码的图像和包含提示、种子、尺寸等元数据的JSON文件。数据集支持通过WebDataset库进行加载和使用，适用于训练和流式处理。

The ZImage-Turbo WebDataset is a collection of generated images from the ZImage-Turbo model using prompts from DiffusionDB. The dataset contains 211,081 samples, divided into 216 shards with approximately 1,000 samples per shard. The generation process utilized 8x NVIDIA RTX 3090 GPUs, took approximately 2 days, and cost around $70. The dataset offers various configurations of base resolutions and aspect ratios, with each shard containing JPEG-XL encoded images and JSON files with metadata including prompts, seeds, dimensions, etc. The dataset supports loading and usage via the WebDataset library, making it suitable for training and streaming purposes.