Datasets:

tandevllc
/

cybersecurity-wiki-slices

license: other
task_categories:
  - text-generation
  - summarization
  - question-answering
  - text-retrieval
  - text-classification
  - sentence-similarity
  - feature-extraction
language:
  - en
tags:
  - wikipedia
  - cybersecurity
  - cyber-security
  - security
  - red-team
  - blue-team
  - penetration-testing
  - pentesting
  - os
  - networking
  - protocols
  - cryptography
  - malware
  - dfir
  - cloud-security
  - threat-intel
  - kubernetes
  - linux
  - windows
  - iot
  - active-directory
  - ics-scada
  - devsecops
  - siem
  - ids-ips
  - dataset
  - jsonl
  - reverse-engineering
pretty_name: Cybersecurity Wiki Slices
size_categories:
  - 10K<n<100K
configs:
  - config_name: wiki
    data_files:
      - split: train
        path: wiki/train-*
dataset_info:
  config_name: wiki
  features:
    - name: text
      dtype: string
    - name: meta
      struct:
        - name: source
          dtype: string
        - name: title
          dtype: string
        - name: pageid
          dtype: int64
        - name: rev_id
          dtype: int64
        - name: source_url
          dtype: string
        - name: license
          dtype: string
        - name: license_url
          dtype: string
        - name: license_class
          dtype: string
        - name: attribution
          dtype: string
        - name: topic
          dtype: string
        - name: redpajama_set_name
          dtype: string
  splits:
    - name: train
      num_bytes: 117597056
      num_examples: 17902
  download_size: 63444989
  dataset_size: 117597056

Cybersecurity Wiki Slices (Parquet)

Cybersecurity Wiki Slices is a curated collection of English Wikipedia pages covering cybersecurity foundations, systems, protocols, tools, and defensive/attack techniques—consolidated into Parquet for fast streaming with the datasets library, with original JSONL preserved under raw/.

Token count: ~24.73M tokens.

⚠️ Ethical use & attribution required. Content originates from Wikipedia and is licensed under CC BY‑SA 4.0. Derivatives must preserve attribution and ShareAlike terms. See Licensing & Attribution below.

What’s in this release (Parquet)

Primary delivery = Parquet shards under data/wiki/.../train-*.parquet (Hub‑managed) for efficient loading.
Raw JSONL retained at raw/wiki.jsonl[.gz|.zst] for transparency and reproducibility.
Named config: wiki (single config; no all config).

Scope & Coverage

This corpus targets practical security knowledge across the stack:

Foundations & Methodologies — computer security, pentesting, vulnerability assessment, OSINT.
Web & API Security — OWASP classes (XSS/SQLi/CSRF/SSRF/etc.), authentication & authorization, CORS/CSP, browser security.
Networks & Protocols — TCP/IP layers, TLS/SSH/QUIC/HTTP/2–3, VPNs, wireless & Bluetooth, MITM classes.
Operating Systems & Internals — Windows internals, PowerShell, WMI/DCOM, Active Directory; Linux subsystems (SELinux, AppArmor, namespaces, cgroups).
Cloud & Containers — AWS/Azure/GCP primitives, IAM, logging/audit, Kubernetes/containers & hardening.
Privilege Escalation, Lateral Movement, Persistence, Credential Access — AD/Windows/Linux concepts and primitives.
Malware & C2 (conceptual) — families, botnets/C2, ransomware; evasion/obfuscation (non‑instructional, encyclopedic POV).
Reverse Engineering & Exploit Dev (conceptual) — RE tooling, memory safety classes, fuzzing.
Crypto & Passwords — algorithms, PKI, KDFs, steganography (encyclopedic coverage).
DFIR / Threat Intel / Logging — IR workflows, indicators, STIX/TAXII, SIEM/log management, IDS/IPS/honeypots.
ICS/SCADA, IoT, Mobile, Automotive — protocols and security considerations.
Supply Chain & DevSecOps — SBOM, signing, CI/CD security; risk/governance and baselines.

The dataset focuses on encyclopedic text. It does not include exploit POCs or instructional “how‑to hack” guides.

Repository Layout

/                                # dataset root (this card lives here as README.md)
  raw/
    wiki.jsonl                   # original JSONL export (optionally .gz/.zst)
  data/
    wiki/
      default/1.0.0/train/
        train-00000-of-XXXXX.parquet
        train-00001-of-XXXXX.parquet
        ...

Each JSONL line (in raw/) or Parquet row (in data/wiki/...) is a single record.

Schema

Core fields

text (string) — cleaned encyclopedic text (normalized whitespace; multi‑paragraph).
meta (object) — metadata with at least:
- source ("wikipedia")
- title (string) — page title
- pageid (int) — Wikipedia page ID
- rev_id (int) — revision id used for extraction
- source_url (string) — canonical page URL
- license ("CC BY-SA 4.0")
- license_url (string) — https://creativecommons.org/licenses/by-sa/4.0/
- license_class ("sharealike")
- attribution (string) — attribution sentence for downstream use
- topic (string) — high‑level slice label (e.g., protocols, active_directory, kubernetes_security, …)
- redpajama_set_name ("WikiSecuritySlices")

Downstream readers should tolerate missing optional keys and nulls.

Loading Examples

1) Load the Parquet config (recommended)

from datasets import load_dataset

REPO = "tandevllc/cybersecurity-wiki-slices"
wiki = load_dataset(REPO, name="wiki", split="train")
len(wiki), wiki.column_names

2) Direct JSONL loading from `raw/` (explicit)

from datasets import load_dataset

wiki = load_dataset(
    "json",
    data_files="raw/wiki.jsonl",
    repo_id="tandevllc/cybersecurity-wiki-slices",
    split="train",
)

3) Filtering by topic or domain

# Keep only networking/protocol pages
net = wiki.filter(lambda r: (r.get("meta") or {}).get("topic") in {"networks", "network_protocols", "protocols"})

# Example: filter by title keyword
tls_pages = wiki.filter(lambda r: "TLS" in (r.get("meta") or {}).get("title", ""))

Cleaning & Quality (High‑Level)

Encyclopedic text blocks only (intro + sections), no wikitext/markup; boilerplate minimized.
Normalization of line breaks and whitespace; removal of control characters.
Minimum‑length guardrails to avoid stubs; duplicates avoided via pageid/revision tracking.
Topical coverage is intentionally broad; some overlap across topics may remain for completeness.

Intended Uses

Pretraining / Continued Pretraining for models needing strong general security background.
RAG/Retrieval over security concepts, protocols, tools, and platform internals.
Evaluation: long‑form QA, summarization, topical recall across security domains.
Education: safe, encyclopedic overviews for bootcamps and courseware.

Not a vulnerability feed; pair with CVE/NVD datasets for up‑to‑date vulnerability intel.

Limitations & Caveats

License & ShareAlike — CC BY‑SA 4.0 requires attribution and sharing derivatives under the same license.
Temporal Drift — Wikipedia pages evolve; this snapshot reflects a specific revision set.
Extraction Noise — Edge cases may include navigation remnants or short stubs.
Topic Labels — meta.topic is a helpful slice label, not a strict taxonomy.

Citation

@dataset{tandevllc_2025_cybersecurity_wiki_slices,
  author = {Gupta, Smridh},
  title  = {Cybersecurity Wiki Slices},
  year   = {2025},
  url    = {https://huggingface.co/datasets/tandevllc/cybersecurity-wiki-slices}
}

Maintainer

Smridh Gupta — smridh@tandev.us