jsantillana/vectrayx-vision-dataset

VectraYX Vision Dataset (CC-BY-SA subset)

A cybersecurity Vision-Language dataset in Spanish, generated from Stack Exchange communities (Security, Cryptography, Reverse Engineering) under CC-BY-SA 4.0.

Dataset Description

12,028 records of (synthetic image, Spanish question, Spanish answer) triples for technical cybersecurity topics.

Data fields

Field	Type	Description
question	string	Technical question in Spanish
answer	string	Detailed answer in Spanish
image_desc	string	Textual description of the synthetic image
image_template	string	PIL template used to render the image
domain	string	soc or re
source_attribution	string	Stack Exchange post attribution
source_url	string	Original Stack Exchange URL
license_id	string	cc-by-sa-4.0

Domain breakdown

Domain	Records	Source
soc (Security Operations)	6,953	Stack Exchange Security
re (Reverse Engineering)	5,075	Stack Exchange Reverse Engineering

Image templates

Synthetic PIL renders (no real screenshots) using templates such as: siem_dashboard, wireshark_packets, log_table, cve_card, terminal_ansi, ida_disasm, ghidra_decomp, gdb_session, hex_dump, register_panel

Images are in the images/ directory organized by domain.

License

• Text data (questions, answers): CC-BY-SA 4.0, derived from Stack Exchange
• Synthetic images: CC-BY-SA 4.0 (derived work — image content generated from CC-BY-SA source text)

Attribution: Stack Exchange community contributors (see source_attribution field).

Usage

This dataset was used to fine-tune VectraYX Vision, a 4B multimodal model for cybersecurity in Latin American Spanish.

from datasets import load_dataset
ds = load_dataset("jsantillana/vectrayx-vision-dataset")