fal/cosmos-imagenet

Tiny Cosmos-Tokenized Imagenet数据集是一个使用Cosmos-tokenized技术压缩的Imagenet数据集，整个数据集仅占用2.45GB的空间。该数据集适用于快速原型开发，提供了连续和离散两种tokenizer的使用方法，并详细说明了如何加载和解码数据。

Tiny Cosmos-Tokenized Imagenet is a dataset based on Cosmos-tokenized Imagenet, designed for rapid prototyping. Using the Cosmos-tokenizer, this dataset is able to compress the entire Imagenet into just 2.45 GB of data, significantly reducing storage requirements. The dataset is provided in pytorch safetensor format, supporting both continuous and discrete tokenizer types, making it suitable for scenarios requiring efficient processing and storage of large-scale image data.