jsantillana/vectrayx-vision-dataset

VectraYX Vision数据集（CC-BY-SA子集）是一个网络安全视觉-语言数据集，主要使用西班牙语，数据来源于Stack Exchange社区（如Security、Cryptography、Reverse Engineering），遵循CC-BY-SA 4.0许可。该数据集包含12,028条记录，每条记录由（合成图像、西班牙语问题、西班牙语答案）三元组构成，涵盖技术性网络安全主题。数据字段包括西班牙语技术问题、详细答案、合成图像的文本描述、用于渲染图像的PIL模板、领域（soc或re）、来源归属、原始Stack Exchange URL和许可证ID。领域分为soc（安全操作，源自Stack Exchange Security）和re（逆向工程，源自Stack Exchange Reverse Engineering）。图像为合成PIL渲染，无真实截图，使用多种模板如siem_dashboard、wireshark_packets等。数据集用于微调VectraYX Vision模型，这是一个针对拉丁美洲西班牙语的4B参数多模态网络安全模型。VectraYX Vision Dataset (CC-BY-SA subset) is a cybersecurity vision-language dataset in Spanish, generated from Stack Exchange communities (Security, Cryptography, Reverse Engineering) under CC-BY-SA 4.0. It contains 12,028 records of (synthetic image, Spanish question, Spanish answer) triples for technical cybersecurity topics. Data fields include technical question in Spanish, detailed answer in Spanish, textual description of the synthetic image, PIL template used to render the image, domain (soc or re), source attribution, original Stack Exchange URL, and license ID. The domain breakdown includes soc (Security Operations, from Stack Exchange Security) and re (Reverse Engineering, from Stack Exchange Reverse Engineering). Images are synthetic PIL renders (no real screenshots) using templates such as siem_dashboard, wireshark_packets, etc. The dataset was used to fine-tune VectraYX Vision, a 4B multimodal model for cybersecurity in Latin American Spanish.