sensenova/SenseNova-U1-Training-Sample

SenseNova-U1训练样本是一个小型示例数据集，旨在配合SenseNova-U1训练代码使用。该数据集仅用于端到端测试训练管道——每个在training/configs/中定义的任务类型都由少量样本表示，以确保数据加载器、打包和损失代码路径都能执行。它不是一个可用于实际训练的训练语料库。数据集包含多种任务类型的样本，覆盖文本到图像（text-to-image）、图像到图像（image-to-image）、图像文本到文本（image-text-to-text）和视觉问答（visual-question-answering）等任务。数据布局包括样本目录（包含JSONL注释文件，每个文件对应不同任务类型，如t2i、it2i、interleave_gen等）、图像目录（包含文本到图像目标、图像编辑源/目标对、多输入图像编辑等子目录）和视频目录（用于视频理解的短视频片段）。数据集总大小约为250个文件/680 MB，支持英语和中文语言，并遵循Apache-2.0许可。数据模式包括顶层元数据JSON文件（如sample_data_meta.json）和每个样本的JSONL注释文件，后者遵循对话格式。数据集还提供了任务覆盖表，详细说明了每个任务类型对应的JSONL文件和相关注释。

SenseNova-U1 Training Sample is a tiny illustrative dataset accompanying the SenseNova-U1 training code. It exists only to smoke-test the training pipeline end-to-end — every task type defined in training/configs/ is represented by a handful of samples so the dataloader / packing / loss code paths all execute. It is not a usable training corpus. The dataset includes samples for multiple task types, covering text-to-image, image-to-image, image-text-to-text, and visual-question-answering tasks. The layout consists of a sample directory (containing JSONL annotation files for different task types, such as t2i, it2i, interleave_gen, etc.), an images directory (with subdirectories for text-to-image targets, image-editing source/target pairs, multi-input image editing, etc.), and a videos directory (with short video clips for video understanding). The total size is approximately 250 files / 680 MB, supporting English and Chinese languages, and licensed under Apache-2.0. The data schema includes a top-level meta JSON file (e.g., sample_data_meta.json) and per-sample JSONL annotation files following a conversational format. The dataset also provides a task coverage table detailing the JSONL files and notes for each task type.