Datasetpublic

SkywardNomad92/smart-contract-audit-findings

Smart Contract Audit Findings Dataset A dataset of 49,611 smart contract security audit findings for fine-tuning LLMs on vulnerability detection. Dataset Description This dataset contains real security audit findings from 30 professional audit firms including Code4rena, OpenZeppelin, Sherlock, Cantina, and others. Formatted for training models to analyze smart contract code and identify vulnerabilities. Splits Split Examples Description… See the full description on the dataset page: https://huggingface.co/datasets/SkywardNomad92/smart-contract-audit-findings.

sourceHugging Facemitupdated 8mo agoView on Hugging Face

1likes84downloads

Dataset Card

Smart Contract Audit Findings Dataset

A dataset of 49,611 smart contract security audit findings for fine-tuning LLMs on vulnerability detection.

Dataset Description

This dataset contains real security audit findings from 30 professional audit firms including Code4rena, OpenZeppelin, Sherlock, Cantina, and others. Formatted for training models to analyze smart contract code and identify vulnerabilities.

Splits

Split	Examples	Description
train	47,130	Training data
validation	2,481	Validation data

Severity Distribution

Severity	Count	Percentage
High	~7,950	16.0%
Medium	~13,686	27.5%
Low	~24,794	49.8%
Gas	~3,331	6.7%

Format

Each example contains a messages array with user/assistant turns:

json

{
  "messages": [
    {
      "role": "user",
      "content": "Analyze this smart contract code for security vulnerabilities:\n\n```solidity\n[code]\n```\n\nProvide a detailed security analysis..."
    },
    {
      "role": "assistant",
      "content": "## [Finding Title]\n\n**Severity**: HIGH Risk\n\n[Detailed analysis with description, impact, and recommendations]"
    }
  ]
}

Usage

python

from datasets import load_dataset

dataset = load_dataset("SkywardNomad92/smart-contract-audit-findings")

Intended Use

—Fine-tuning LLMs for smart contract security analysis
—Training vulnerability detection models
—Domain adaptation for blockchain security