extract 0 0 0

Blainer28/forge-reason-v1

published by @Blainer28

Alignment dataset tracked in the /datasets sub-catalog.

Lifetime downloads
159
Last 30 days
17
Likes
-
Size
626 KB
Created on HF
2026-04-13
Age
6 months ago

Description

FORGE-REASON v1

Dataset Description

FORGE-REASON is the first open-source dataset of red-teamed mathematical proofs

for fine-tuning LLMs on formal logical reasoning. Each entry contains a triple:

flawed_proof → directive4_critique → corrected_proof

Proofs span four mathematical domains: computational complexity theory, number theory,

cryptography (protocol security), and combinatorics.

Intended Use

Fine-tuning open-source LLMs (Llama, Mistral… See the full description on the dataset page: https://huggingface.co/datasets/Blainer28/forge-reason-v1.

Tags

task_categories:text-generationtask_categories:question-answeringlanguage:enlicense:apache-2.0size_categories:n<1Kformat:parquetformat:optimized-parquetmodality:tabularmodality:textlibrary:datasetslibrary:pandaslibrary:polarslibrary:mlcroissantregion:usmathematicsreasoningred-teamingformal-verificationcomplexity-theorysynthetic

README current version from Hugging Face


license: apache-2.0
task_categories:

  • text-generation
  • question-answering
    language:
  • en
    tags:
  • mathematics
  • reasoning
  • red-teaming
  • formal-verification
  • complexity-theory
  • synthetic
    pretty_name: FORGE-REASON v1
    size_categories:
  • 1K<n<10K

FORGE-REASON v1

Dataset Description

FORGE-REASON is the first open-source dataset of red-teamed mathematical proofs
for fine-tuning LLMs on formal logical reasoning. Each entry contains a triple:

flawed_proof → directive4_critique → corrected_proof

Proofs span four mathematical domains: computational complexity theory, number theory,
cryptography (protocol security), and combinatorics.

Intended Use

  • Fine-tuning open-source LLMs (Llama, Mistral, Qwen) on mathematical reasoning
  • Training hostile peer-reviewer models (Directive-4 style)
  • Benchmarking LLM robustness to subtle logical flaws
  • Research on formal verification of natural language proofs

Out-of-Scope Use

  • This dataset contains intentionally flawed proofs. Do NOT use flawed_proof entries
    as ground-truth mathematics. Always use the corrected_proof field for valid content.
  • Not suitable for formal theorem provers without human review.

Dataset Statistics

Split Entries
train ~80%
validation ~10%
test ~10%

Domains: complexity, number_theory, cryptography, combinatorics
Difficulty: graduate-level
Flaw types: circular_dependency, domain_restriction, hidden_assumption,
overclaim, missing_base_case, invalid_reduction

Example Entry

{
  "id": "forge-reason-00001",
  "domain": "complexity",
  "difficulty": "graduate",
  "flawed_proof": {
    "claim": "P = NP via polynomial-time 3-SAT solver",
    "proof_text": "Let A be a 3-SAT instance...",
    "flaw_type": "circular_dependency",
    "flaw_description": "Reduction assumes NP ⊆ P to invoke the oracle"
  },
  "critique": {
    "directive4_report": "Step 3 invokes an NP oracle...",
    "identified_flaw_type": "circular_dependency",
    "confidence": 0.92,
    "findings": ["Oracle invocation at step 3 assumes P=NP"]
  },
  "correction": {
    "corrected_proof": "Restrict to class of bounded-depth circuits...",
    "correction_type": "scope_restriction",
    "explanation": "The claim is restricted to NC¹ circuits where..."
  },
  "metadata": {
    "z3_validated": false,
    "generation_model": "claude-sonnet-4-6",
    "critique_model": "claude-sonnet-4-6",
    "timestamp": "2026-04-13T00:00:00+00:00",
    "tokens_used": 1842
  }
}

Contact

Dominik Blain — [email protected]

Inquiries about custom dataset generation, fine-tuning engagements,
or research collaborations welcome.

License

Apache 2.0

Citation

@dataset{forge_reason_v1_2026,
  title={FORGE-REASON v1: Synthetic Red-Teamed Mathematical Proofs},
  author={Blain, Dominik},
  year={2026},
  url={https://huggingface.co/datasets/Blainer28/forge-reason-v1}
}

README history 3 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. 2026-04-13Add contact info to dataset carde582eb02.9 KB
    Loading...
  2. 2026-04-13Upload datasetd7340851.2 KB
    Loading...
  3. 2026-04-13initial commitc40792c0 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.

Community discussions 1 thread · 1 open · 1 comments

Threads posted by the community on Hugging Face: bug reports, questions about usage, pull requests on the README, license clarifications. Click any thread to expand and read the full conversation inline. Newest first.

  1. 2026-04-14[bot] Conversion to Parquetopen1 💬#1
    Loading...

Related datasets same stage · extract

Read further

Metadata

License
apache-2.0

Repository

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("Blainer28/forge-reason-v1")
Raw snapshot (Python)
from huggingface_hub import snapshot_download
snapshot_download(repo_id="Blainer28/forge-reason-v1", repo_type="dataset")
git clone (requires git-lfs)
git clone https://huggingface.co/datasets/Blainer28/forge-reason-v1
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