LLM Datasets

This dataset contains 138,172 video clips covering diverse real-world scenes and entity-interaction relationships. Each video clip includes keyframes that represent important entity interactions and their corresponding relationships. The annotation files provide entity types, bounding boxes, and interaction labels for relationship triples identified in the keyframes. Each relationship can be represented as a structured triple: (entity1, entity2, interaction). This dataset can be used for video understanding task.

This dataset focuses on the four major European languages (French, German, Spanish, Italian) and contains 480000 pairs of original and corrected text pairs. Each piece of data is presented in JSON format, including two fields: input (raw text) and output (corrected text), which can assist in natural language processing, machine translation, and language teaching research.

This dataset is designed to tackle the core weaknesses of today's large language models when it comes to processing long documents and performing complex reasoning. It consists of 7,500 high-quality training examples across three languages—Chinese, English, and Korean. Each instance is built around a long-text passage and includes questions that require synthesizing information across paragraphs and documents, while following multi-step logical chains. The goal is to offer a thorough and rigorous evaluation framework that tests a model's ability to perceive long-range context, retrieve relevant information, construct sound reasoning paths, and trace evidence back to its source.

This dataset covers office-based scenarios such as in-depth searches, data analysis, and industry research, encompassing complete multi-turn reasoning trajectories and tool-calling chains. It is designed to support the analysis of agent planning capabilities, research into tool selection strategies, and quality assessment, providing a structured benchmark for agent training and evaluation.

2This dataset contains 202,735 presentation slide images. The dataset covers four major visual types commonly found in presentation slides: structural diagrams, graphs, flowcharts, and figures. The main text of the slides is available in either Chinese or English. Each annotation file is labeled according to the language of the corresponding presentation slide. This dataset can be used in document intelligence, document understanding, visual question answering (VQA), and vision-language model training.

This dataset contains 204,522 test paper questions, covering multiple subjects, question types, and collection devices (mobile phones, scanners). The text is fully transcribed, and formulas and tables are transcribed using LaTeX format. This dataset can be used for tasks such as intelligent exam paper marking and homework tutoring. We strictly adhere to data protection regulations and privacy standards, ensuring the maintenance of user privacy and legal rights throughout the data collection, storage, and usage processes, our datasets are all GDPR, CCPA, PIPL complied.

1,400 Human Action Image Dataset with 5,937 Boxes Annotate Data collected a variety of scenes and human activities. Each person in the image is annotated with detailed descriptions.This data can provide a rich resource for large multi-modal models. It has been validated by multiple AI companies and proves beneficial for achieving outstanding performance in real-world applications. Throughout the process of Dataset collection, storage, and usage, we have consistently adhered to dataset protection and privacy regulations to ensure the preservation of user privacy and legal rights. All Dataset comply with regulations such as GDPR, CCPA, PIPL, and other applicable laws.

This dataset consisting of 20,846 recipe groups. Each recipe contains between 4-18 images and each image with a corresponding textual description. Cuisines include Chinese, Western, Korean, Japanese and others. Description languages are Chinese and English, with minimum lengths of 15 Chinese words and 30 English words per description. The data is suitable for vision language model training, image captioning, multimodal recipe understanding, and instruction-following tasks related to cooking workflows.

This dataset contains 100,145 annotated icon samples. The data includes two major categories of icons, 3D Style Icons and Vector Illustration Icons, covering a total of 16 subcategories. In terms of annotation, Each icon is paired with a concise Chinese textual description of about 30 characters. The data can be used for tasks such icon recognition, vision–language model training, user interface understanding.