extract 0 0 0

fgnfh/Frankestain

published by @fgnfh

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
67
Last 30 days
16
Likes
-
Size
262 KB
Created on HF
2026-09-09
Age
1 months ago

Description

Frankestain

Private instruction-tuning dataset for a personal cybersecurity study assistant aligned to a phased learning roadmap (Linux/networking → scripting → low-level → structured paths).

Authorized lab use only. Content is for learning on systems you own or have explicit permission to test. Do not use it to attack networks, devices, or accounts you do not own.

Dataset summary

Format: JSON array of objects

Fields:

instruction (string) — user/ask prompt

output… See the full description on the dataset page: https://huggingface.co/datasets/fgnfh/Frankestain.

Tags

task_categories:text-generationtask_categories:question-answeringlanguage:enlanguage:itlicense:othersize_categories:n<1Kregion:uscybersecurityblue-teamred-teamlinuxwindowsauthorized-lab-only

README current version from Hugging Face


license: other
task_categories:

  • text-generation
  • question-answering
    language:
  • en
  • it
    tags:
  • cybersecurity
  • blue-team
  • red-team
  • linux
  • windows
  • authorized-lab-only
    pretty_name: Frankestain
    size_categories:
  • n<1K

Frankestain

Private instruction-tuning dataset for a personal cybersecurity study assistant aligned to a phased learning roadmap (Linux/networking → scripting → low-level → structured paths).
Authorized lab use only. Content is for learning on systems you own or have explicit permission to test. Do not use it to attack networks, devices, or accounts you do not own.

Dataset summary

  • Format: JSON array of objects
  • Fields:
    • instruction (string) — user/ask prompt
    • output (string) — target assistant reply
    • topic (string) — content bucket (e.g. bandit_system_skills, linux_permissions_and_files, duckyscript_advanced)
    • phase (string) — roadmap phase 1–4
    • side (string) — blue or red framing
  • Language mix: Italian and English prompts/answers appear in the data
  • Scale: hundreds of examples (small / growing); not a production-scale corpus

Phases (roadmap alignment)

Phase Focus (high level)
1 Linux basics: shell, files, permissions, processes, Bandit-style skills
2 Networking / scripting-oriented security topics
3 Lower-level and specialized topics (e.g. analysis, HID/DuckyScript lab material)
4 Broader structured-path style material
Topic labels should match the phase. Phase 1 is reserved for foundational Linux skills, not advanced Windows exploitation.

Intended use

  • Supervised fine-tuning (SFT) / LoRA of a small assistant that explains concepts and lab workflows
  • Filtering by phase, topic, or side for curriculum-specific adapters
  • Holding out an evaluation split (do not train on every row)

Out of scope

  • Not a substitute for hands-on practice (terminal, labs, hardware you own)
  • Not a source of “copy-paste attack” playbooks for unauthorized use
  • Not sufficient alone for strong code generation (e.g. C) or expert DuckyScript without many more dedicated examples and hardware validation

Data quality notes

  • Prefer diverse, non-duplicated examples over near-copies
  • Keep phase / topic consistent with difficulty
  • Document lab-only assumptions in answers where actions are described

File layout

  • merged_dataset_all.json — main training file
  • README.md — this card

Citation / ownership

Private dataset maintained by the repo owner for personal education. Redistribution only as allowed by the owner.

README history 8 revisions

Every night we snapshot the README of every dataset in the catalog. When the SHA changes we archive the new version and diff it against the last. This is the evolving thought record of the alignment-data field: what the author decided to say about the corpus, and how that framing shifted over time.

  1. 2026-09-09Upload 0.55b379e42.6 KB
    Loading...
  2. 2026-09-09Delete v0.4.jsond7bf6862.6 KB
    Loading...
  3. 2026-09-09Update README.md1bf5b492.6 KB
    Loading...
  4. 2026-09-09Delete merged_dataset_all.jsonc9be1a50 B
    Loading...
  5. 2026-09-09Upload v0.4.jsonfa707600 B
    Loading...
  6. 2026-09-09Upload merged_dataset_all.jsonf6f0fa90 B
    Loading...
  7. 2026-09-09Create README.mdd7892cd0 B
    Loading...
  8. 2026-09-09initial commit31c4ace0 B
    Loading...

Used by abliterated models none yet

No abliterated models in our catalog have declared this dataset in their YAML metadata yet. This can mean the dataset is used but not declared, is used outside abliteration workflows, or is new. The link will populate automatically as models are indexed.

Related datasets same stage · extract

Read further

Metadata

License
other

Download the dataset

0

Copy any of these snippets into your notebook or terminal. All three fetch directly from Hugging Face using your own credentials.

Recommended: datasets library (Python)
from datasets import load_dataset
ds = load_dataset("fgnfh/Frankestain")
Raw snapshot (Python)
from huggingface_hub import snapshot_download
snapshot_download(repo_id="fgnfh/Frankestain", repo_type="dataset")
git clone (requires git-lfs)
git clone https://huggingface.co/datasets/fgnfh/Frankestain
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration