Datasets:
| license: apache-2.0 | |
| task_categories: | |
| - text-generation | |
| - text2text-generation | |
| language: | |
| - en | |
| - code | |
| tags: | |
| - cybersecurity | |
| - vulnerability-detection | |
| - cve | |
| - code-audit | |
| - code-repair | |
| size_categories: | |
| - 1M<n<10M | |
| # FineSec-1M Master Security Dataset | |
| **FineSec-1M** is a high-precision, 1.11 GB cybersecurity dataset containing **1,000,000 training examples** (500,000 vulnerable code snippets + 500,000 safe control examples) designed for fine-tuning state-of-the-art security LLMs. | |
| ## Dataset Structure | |
| 1. **Server-Side Request Forgery (SSRF - CWE-918)**: 167,054 examples | |
| 2. **Unquoted Search Paths (CWE-428)**: 166,525 examples | |
| 3. **Insecure Randomness (CWE-330)**: 166,421 examples | |
| 4. **Safe Control Code Suite (N/A)**: 500,000 examples | |
| ## Class & Language Balance | |
| - **Class Balance**: Exactly 50.0% Vulnerable | 50.0% Safe Controls (Guarantees zero false positives). | |
| - **Languages**: Python (66.7%) & C/C++ (33.3%). | |
| ## Usage with Hugging Face Datasets | |
| ```python | |
| from datasets import load_dataset | |
| dataset = load_dataset('elsiddik/zz', split='train') | |
| print(f'Total records: {len(dataset)}') | |
| ``` | |