sensenova/SenseNova-SI-8M

SenseNova-SI-8M是一个用于空间智能研究的官方大规模训练数据集，包含约816万个精心策划的训练样本，覆盖约272万张唯一图像，基于严格的空间能力分类法组织。该数据集用于训练推荐发布的模型SenseNova-SI-1.1-InternVL3-8B，并作为SenseNova-SI系列中空间智能研究的规范训练语料库。数据以JSONL格式存储，每个条目包含唯一标识符id、对话列表conversations和图像路径列表image，其中conversations中的<image>占位符指示图像插入位置。数据集旨在通过多样化数据训练提升多模态基础模型的空间智能能力，支持评估和下游应用。

SenseNova-SI-8M is the official full-scale training dataset for spatial intelligence research, containing approximately 8.16 million carefully curated training samples spanning about 2.72 million unique images, organized under a rigorous taxonomy of spatial capabilities. It is used to train the recommended released model SenseNova-SI-1.1-InternVL3-8B and serves as the canonical training corpus for the SenseNova-SI series in spatial intelligence studies. The data is stored in JSONL format, with each entry including a unique identifier id, a conversations list of dialogue turns, and an image list of file paths, where <image> placeholders in conversations indicate image insertion points. The dataset aims to enhance spatial intelligence in multimodal foundation models through diverse data training, supporting evaluation and downstream applications.