Spaces:
Running
Running
metadata
title: SafeLLM Leaderboard
emoji: 🛡️
colorFrom: indigo
colorTo: blue
sdk: docker
pinned: true
license: apache-2.0
short_description: Trusted OSS Model Supply Chain Security Rankings
🏰 SafeLLM Leaderboard
Comprehensive security rankings for machine learning models
📊 About
This leaderboard displays security rankings for ML models scanned with Palisade, a comprehensive security scanner that detects:
- 🎯 Backdoors & Trojans - Hidden malicious behaviors
- 🔓 Pickle RCE - Remote code execution vulnerabilities
- 💥 Buffer Overflows - Memory safety issues
- 🔗 Supply Chain Attacks - Compromised dependencies
- 🔐 Model Integrity - Tampering detection
- 🎭 Tokenizer Hijacking - Malicious configurations
🎯 Understanding the Scores
Security Score
Lower is better! Calculated as:
Score = (Critical × 100) + (High × 50) + (Medium × 10) + (Low × 2)
- 0-49: ✅ Excellent security
- 50-99: 🟡 Good security
- 100-199: 🟠 Moderate concerns
- 200+: 🔴 Significant issues
Risk Levels
| Level | Meaning | Action |
|---|---|---|
| 🟢 Safe | No significant issues | Deploy with confidence |
| 🟡 Low | Minor issues only | Review and monitor |
| 🟠 Medium | Some concerns | Fix before production |
| 🔴 High | Serious issues | Use with caution |
| ⛔ Critical | Critical vulnerabilities | Do NOT use |
📈 Features
- Interactive Filtering - By risk level, score, and size
- Rich Visualizations - Charts and graphs powered by Plotly
- Detailed Analysis - Threat categories and MITRE ATT&CK mapping
- SARIF Reports - Industry-standard security reports
- Real-time Updates - Auto-refreshes from HuggingFace dataset
🔍 Data Source
All scan results are stored in the public dataset: javelinai/palisade-scan-results
Models are scanned weekly with automated GitHub Actions.
🛠️ Technology Stack
- Scanner: Palisade
- Frontend: Gradio 4.27
- Visualizations: Plotly
- Data: HuggingFace Datasets
- Hosting: HuggingFace Spaces