Upload README.md with huggingface_hub

Browse files

Files changed (1) hide show

README.md +132 -629

README.md CHANGED Viewed

@@ -1,704 +1,207 @@
 ---
-license: apache-2.0
 base_model: google/codegemma-7b-it
 tags:
-- code
-- security
-- codegemma
-- google
-- securecode
-- owasp
-- vulnerability-detection
 datasets:
-- scthornton/securecode-v2
-language:
-- en
-library_name: transformers
 pipeline_tag: text-generation
-arxiv: 2512.18542
 ---
-# CodeGemma 7B - SecureCode Edition
 <div align="center">
-[![License](https://img.shields.io/badge/license-Apache%202.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
-[![Training Dataset](https://img.shields.io/badge/dataset-SecureCode%20v2.0-green.svg)](https://huggingface.co/datasets/scthornton/securecode-v2)
-[![Base Model](https://img.shields.io/badge/base-CodeGemma%207B-orange.svg)](https://huggingface.co/google/codegemma-7b-it)
-[![perfecXion.ai](https://img.shields.io/badge/by-perfecXion.ai-purple.svg)](https://perfecxion.ai)
-**🔷 Google's code model enhanced with security expertise**
-Exceptional instruction following meets security awareness. Perfect for developers who want Google's proven quality with security-first coding.
-[📄 Paper](https://arxiv.org/abs/2512.18542) | [🤗 Model Hub](https://huggingface.co/scthornton/codegemma-7b-securecode) | [📊 Dataset](https://huggingface.co/datasets/scthornton/securecode-v2) | [💻 perfecXion.ai](https://perfecxion.ai) | [📚 Collection](https://huggingface.co/collections/scthornton/securecode)
 </div>
 ---
-## 🎯 Quick Decision Guide
-**Choose This Model If:**
-- ✅ You value **Google brand trust** and proven quality
-- ✅ You need **excellent instruction following** for complex security tasks
-- ✅ You want **strong code completion** with security awareness
-- ✅ You're building on **Google Cloud Platform** or Google ecosystem
-- ✅ You need **reliable, consistent responses** from a proven architecture
-- ✅ You prefer **7B efficiency** with Google's engineering quality
-**Consider Other Models If:**
-- ⚠️ You need maximum context window (→ Qwen 7B/14B with 128K)
-- ⚠️ You're on very limited hardware (→ Llama 3B)
-- ⚠️ You need enterprise brand diversity (→ IBM Granite, Meta CodeLlama)
-- ⚠️ You want absolute best code understanding (→ Qwen 7B slightly edges out)
----
-## 📊 Collection Positioning
-| Model | Size | Best For | Hardware | Inference Speed | Unique Strength |
-|-------|------|----------|----------|-----------------|-----------------|
-| Llama 3.2 3B | 3B | Consumer deployment | 8GB RAM | ⚡⚡⚡ Fastest | Most accessible |
-| DeepSeek 6.7B | 6.7B | Security-optimized baseline | 16GB RAM | ⚡⚡ Fast | Security architecture |
-| Qwen 7B | 7B | Best code understanding | 16GB RAM | ⚡⚡ Fast | Best-in-class 7B |
-| **CodeGemma 7B** | **7B** | **Google ecosystem** | **16GB RAM** | **⚡⚡ Fast** | **Instruction following, Google quality** |
-| CodeLlama 13B | 13B | Enterprise trust | 24GB RAM | ⚡ Medium | Meta brand, proven |
-| Qwen 14B | 14B | Advanced analysis | 32GB RAM | ⚡ Medium | 128K context window |
-| StarCoder2 15B | 15B | Multi-language specialist | 32GB RAM | ⚡ Medium | 600+ languages |
-| Granite 20B | 20B | Enterprise-scale | 48GB RAM | Medium | IBM trust, largest |
-**This Model's Sweet Spot:** Google quality + security expertise. Best for teams who value Google's engineering rigor and want proven, reliable security guidance.
----
-## 🚨 The Problem This Solves
-**AI coding assistants produce vulnerable code in 45% of security-relevant scenarios** (Veracode 2025). While many code models focus on syntax and functionality, they lack security awareness.
-**Real-world costs:**
-- **Equifax** (SQL injection): $425 million settlement + brand destruction
-- **Capital One** (SSRF): 100 million customer records, $80M fine
-- **SolarWinds** (authentication bypass): 18,000 organizations compromised
-- **LastPass** (cryptographic failures): 30 million users affected
-CodeGemma SecureCode Edition brings Google's renowned engineering quality to secure coding, combining reliable instruction following with comprehensive security knowledge.
----
-## 💡 What is This?
-This is **Google CodeGemma 7B Instruct** fine-tuned on the **SecureCode v2.0 dataset** - Google's specialized code model enhanced with production-grade security expertise covering the complete OWASP Top 10:2025.
-CodeGemma is part of Google's Gemma family, built on the same technology powering Google's AI products. It's specifically optimized for code generation with exceptional instruction-following capabilities.
-Combined with SecureCode training, this model delivers:
-✅ **Excellent instruction following** - Reliably follows complex security requirements
-✅ **Google engineering quality** - Proven architecture from Google AI
-✅ **Strong code completion** - Exceptional at completing partial secure code
-✅ **Consistent, reliable responses** - Predictable behavior for production use
-✅ **Security-first code generation** - Trained on real vulnerability patterns
-**The Result:** A code assistant that combines Google's quality with security expertise.
-**Why CodeGemma 7B?** This model offers Google's advantages:
-- 🔷 **Google brand trust** - Built by the team behind TensorFlow, BERT, and PaLM
-- 🎯 **Instruction-following excellence** - Consistently follows complex security specifications
-- ⚡ **Production efficiency** - 7B parameters = fast inference
-- 🌍 **Broad language support** - Code generation across major languages
-- 🏢 **GCP integration** - Optimized for Google Cloud Platform deployment
-- ⚖️ **Apache 2.0 licensed** - Full commercial freedom
-Perfect for development teams using Google Cloud, organizations valuing Google's engineering culture, and developers who prioritize instruction-following reliability.
----
-## 🔐 Security Training Coverage
-### Real-World Vulnerability Distribution
-Trained on 1,209 security examples with real CVE grounding:
-| OWASP Category | Examples | Real Incidents |
-|----------------|----------|----------------|
-| **Broken Access Control** | 224 | Equifax, Facebook, Uber |
-| **Authentication Failures** | 199 | SolarWinds, Okta, LastPass |
-| **Injection Attacks** | 125 | Capital One, Yahoo, LinkedIn |
-| **Cryptographic Failures** | 115 | LastPass, Adobe, Dropbox |
-| **Security Misconfiguration** | 98 | Tesla, MongoDB, Elasticsearch |
-| **Vulnerable Components** | 87 | Log4Shell, Heartbleed, Struts |
-| **Identification/Auth Failures** | 84 | Twitter, GitHub, Reddit |
-| **Software/Data Integrity** | 78 | SolarWinds, Codecov, npm |
-| **Logging Failures** | 71 | Various incident responses |
-| **SSRF** | 69 | Capital One, Shopify |
-| **Insecure Design** | 59 | Architectural flaws |
-### Multi-Language Support
-Fine-tuned on security examples across:
-- **Python** (Django, Flask, FastAPI) - 280 examples
-- **JavaScript/TypeScript** (Express, NestJS, React) - 245 examples
-- **Java** (Spring Boot) - 178 examples
-- **Go** (Gin framework) - 145 examples
-- **PHP** (Laravel, Symfony) - 112 examples
-- **C#** (ASP.NET Core) - 89 examples
-- **Ruby** (Rails) - 67 examples
-- **Rust** (Actix, Rocket) - 45 examples
-- **C/C++** (Memory safety) - 28 examples
-- **Kotlin, Swift** - 20 examples
----
-## 🎯 Deployment Scenarios
-### Scenario 1: Google Cloud Platform Integration
-**Native integration with GCP services.**
-**Platform:** Google Cloud Run, Vertex AI, GKE
-**Hardware:** Cloud TPU, NVIDIA T4/A100
-**Use Case:** Serverless security code generation
-**GCP Benefits:**
-- Optimized for Google Cloud infrastructure
-- Seamless Vertex AI integration
-- Cloud Run auto-scaling
-- Integrated monitoring and logging
-**ROI:** Reduced deployment complexity on GCP. Natural fit for Google-first organizations.
----
-### Scenario 2: Secure API Code Generation
-**Generate production-ready secure APIs with precise specifications.**
-**Hardware:** Standard cloud instance (16GB RAM)
-**Use Case:** API security automation
-**Strength:** Follows detailed security requirements precisely
-**Example Use Case:**
-```
-Generate a secure REST API for user authentication with:
-- JWT tokens (RS256)
-- Refresh token rotation
-- Rate limiting (10 req/min per IP)
-- Comprehensive audit logging
-- CSRF protection
-```
-**Instruction Following:** CodeGemma reliably implements ALL specified requirements, not just some.
----
-### Scenario 3: Code Review Copilot
-**Real-time security suggestions during code review.**
-**Platform:** GitHub Copilot alternative, IDE plugins
-**Latency:** <100ms for inline suggestions
-**Use Case:** Security-aware code completion
-**Value Proposition:**
-- Suggests secure patterns as developers type
-- Catches vulnerabilities during development
-- Educates developers on security best practices
-- Reduces security debt accumulation
----
-### Scenario 4: Educational Platform
-**Teaching secure coding with Google-quality foundations.**
-**Audience:** CS students, bootcamp students, junior developers
-**Platform:** Interactive coding platforms
-**Use Case:** Security education at scale
-**Educational Benefits:**
-- Google brand credibility for students
-- Consistent, predictable teaching responses
-- Clear explanations of security concepts
-- Reliable code examples
----
-## 📊 Training Details
-| Parameter | Value | Why This Matters |
-|-----------|-------|------------------|
-| **Base Model** | google/codegemma-7b-it | Google's instruction-tuned code model |
-| **Fine-tuning Method** | LoRA (Low-Rank Adaptation) | Efficient training, preserves base capabilities |
-| **Training Dataset** | [SecureCode v2.0](https://huggingface.co/datasets/scthornton/securecode-v2) | 100% incident-grounded, expert-validated |
-| **Dataset Size** | 841 training examples | Focused on quality over quantity |
-| **Training Epochs** | 3 | Optimal convergence without overfitting |
-| **LoRA Rank (r)** | 16 | Balanced parameter efficiency |
-| **LoRA Alpha** | 32 | Learning rate scaling factor |
-| **Learning Rate** | 2e-4 | Standard for LoRA fine-tuning |
-| **Quantization** | 4-bit (bitsandbytes) | Enables efficient training |
-| **Trainable Parameters** | ~40M (0.57% of 7B total) | Minimal parameters, maximum impact |
-| **Total Parameters** | 7B | Sweet spot for efficiency |
-| **Context Window** | 8K tokens | Standard for code analysis |
-| **GPU Used** | NVIDIA A100 40GB | Enterprise training infrastructure |
-| **Training Time** | ~6 hours (estimated) | Efficient training cycle |
-### Training Methodology
-**LoRA (Low-Rank Adaptation)** preserves CodeGemma's instruction-following capabilities:
-1. **Efficiency:** Trains only 0.57% of model parameters (40M vs 7B)
-2. **Quality:** Maintains Google's exceptional code generation
-3. **Reliability:** Preserves consistent, predictable behavior
-**Google Gemma Foundation:** Built on Google's cutting-edge AI research:
-- State-of-the-art instruction following
-- Optimized for code generation tasks
-- Proven reliability in production
-- Backed by Google AI engineering
----
-## 🚀 Usage
-### Quick Start
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-from peft import PeftModel
-# Load Google CodeGemma base model
-base_model = "google/codegemma-7b-it"
-model = AutoModelForCausalLM.from_pretrained(
-    base_model,
-    device_map="auto",
-    torch_dtype="auto",
-    trust_remote_code=True
-)
-tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
-# Load SecureCode LoRA adapter
-model = PeftModel.from_pretrained(model, "scthornton/codegemma-7b-securecode")
-# Generate secure code with precise requirements
-prompt = """### User:
-Generate a secure user registration endpoint in Python Flask with these exact requirements:
-1. Email validation with regex
-2. Password: minimum 12 chars, complexity requirements
-3. Bcrypt hashing (cost factor 12)
-4. Rate limiting: 5 attempts per 15 minutes per IP
-5. CSRF token validation
-6. SQL injection prevention via parameterized queries
-7. Comprehensive audit logging to Stackdriver
-8. Return JSON with proper status codes
-### Assistant:
-"""
-inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-outputs = model.generate(
-    **inputs,
-    max_new_tokens=2048,
-    temperature=0.7,
-    top_p=0.95,
-    do_sample=True
-)
-response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-print(response)
-```
----
-### GCP Deployment (Vertex AI)
 ```python
-from google.cloud import aiplatform
-from transformers import AutoModelForCausalLM
 from peft import PeftModel
-# Initialize Vertex AI
-aiplatform.init(project='your-project', location='us-central1')
-# Deploy CodeGemma SecureCode to Vertex AI
-model = AutoModelForCausalLM.from_pretrained("google/codegemma-7b-it", device_map="auto")
-model = PeftModel.from_pretrained(model, "scthornton/codegemma-7b-securecode")
-# Upload to Vertex AI Model Registry
-# Deploy as endpoint for production use
-# Integrate with Cloud Run, GKE, or other GCP services
-```
----
-### Production Deployment (4-bit Quantization)
-```python
 from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
-from peft import PeftModel
-# 4-bit quantization - runs on 16GB GPU
 bnb_config = BitsAndBytesConfig(
     load_in_4bit=True,
-    bnb_4bit_use_double_quant=True,
     bnb_4bit_quant_type="nf4",
-    bnb_4bit_compute_dtype="bfloat16"
 )
-model = AutoModelForCausalLM.from_pretrained(
     "google/codegemma-7b-it",
     quantization_config=bnb_config,
     device_map="auto",
-    trust_remote_code=True
 )
-model = PeftModel.from_pretrained(model, "scthornton/codegemma-7b-securecode")
-tokenizer = AutoTokenizer.from_pretrained("google/codegemma-7b-it", trust_remote_code=True)
-# Production-ready: Runs on RTX 3090, RTX 4080, A5000, or GCP T4
 ```
----
-## 📈 Performance & Benchmarks
-### Hardware Requirements
-| Deployment | RAM | GPU VRAM | Tokens/Second | Latency (2K response) | Cost/Month |
-|-----------|-----|----------|---------------|----------------------|------------|
-| **4-bit Quantized** | 16GB | 12GB | ~40 tok/s | ~50 seconds | $0 (local) or $50-100 (cloud) |
-| **8-bit Quantized** | 20GB | 16GB | ~50 tok/s | ~40 seconds | $0 (local) or $100-150 (cloud) |
-| **Full Precision (bf16)** | 28GB | 20GB | ~65 tok/s | ~31 seconds | $0 (local) or $200-300 (cloud) |
-| **GCP Vertex AI** | Managed | Managed | ~60 tok/s | ~33 seconds | $150-250 (pay-per-use) |
-**GCP Integration Winner:** Native Vertex AI deployment with Google's infrastructure optimization.
-### Real-World Performance
-**Tested on RTX 3090 24GB** (consumer/prosumer GPU):
-- **Tokens/second:** ~40 tok/s (4-bit), ~60 tok/s (full precision)
-- **Cold start:** ~3 seconds
-- **Memory usage:** 10GB (4-bit), 16GB (full precision)
-- **Instruction following:** Excellent - implements 95%+ of specified requirements
-**Tested on GCP T4 GPU** (cloud deployment):
-- **Tokens/second:** ~35 tok/s (optimized for cost)
-- **Auto-scaling:** 0 to 100 instances in <60 seconds
-- **Cost efficiency:** $0.35/hour per instance
-### Code Generation Quality
-**Instruction Following Benchmark:**
-- **Requirement compliance:** 95% (implements specified requirements accurately)
-- **Security specification adherence:** Excellent
-- **Consistency:** High - predictable, reliable outputs
----
-## 💰 Cost Analysis
-### Total Cost of Ownership (TCO) - 1 Year
-**Option 1: GCP Vertex AI (Recommended for GCP Users)**
-- Deployment: Managed Vertex AI endpoint
-- Cost: ~$0.50/hour (auto-scaling)
-- Usage: 500 hours/month
-- **Total Year 1:** $3,000/year
-**Option 2: Self-Hosted (Cloud GPU)**
-- GCP n1-highmem-8 + T4 GPU: $0.55/hour
-- Usage: 160 hours/month (development team)
-- **Total Year 1:** $1,056/year
-**Option 3: Self-Hosted (Local GPU)**
-- Hardware: RTX 3090 24GB - $1,000-1,200 (one-time)
-- Electricity: ~$60/year
-- **Total Year 1:** $1,060-1,260
-- **Total Year 2+:** $60/year
-**Option 4: Google Gemini API (for comparison)**
-- Cost: Variable pricing
-- Typical usage: $1,500-3,000/year for team
-- **Total Year 1:** $1,500-3,000/year
-**ROI Winner:** GCP Vertex AI for Google-first orgs (native integration). Local GPU for multi-cloud or cost optimization.
----
-## 🎯 Use Cases & Examples
-### 1. Secure API Generation with Precise Specifications
-Generate APIs that exactly match security requirements:
-```python
-prompt = """### User:
-Create a secure payment processing API endpoint in Node.js/Express with:
-- Input validation using Joi
-- PCI-DSS compliant data handling
-- Stripe integration with webhook verification
-- Idempotency key support
-- Comprehensive error handling
-- Rate limiting (100 req/min)
-- Request/response logging to Stackdriver
-### Assistant:
-"""
-```
-**Model Response:** Generates complete, production-ready code implementing ALL specified requirements.
----
-### 2. Security Code Review with Structured Output
-Review code with predictable, structured responses:
-```python
-prompt = """### User:
-Review this authentication code for OWASP Top 10 vulnerabilities. Provide output in this exact format:
-1. Vulnerability Type
-2. Severity (Critical/High/Medium/Low)
-3. Affected Code Line
-4. Exploitation Scenario
-5. Secure Alternative
-6. OWASP Category
-[Code to review]
-### Assistant:
-"""
-```
-**Model Response:** Follows the exact format specified, reliable structured output.
----
-### 3. Educational Content Generation
-Generate consistent educational examples:
-```python
-prompt = """### User:
-Create a teaching example showing SQL injection vulnerability and fix. Include:
-1. Vulnerable code with clear comments
-2. Attack demonstration
-3. Secure code with parameterized queries
-4. Explanation suitable for beginners
-5. Practice exercise
-### Assistant:
-"""
-```
-**Model Response:** Generates clear, educational content following Google's technical writing standards.
----
-## ⚠️ Limitations & Transparency
-### What This Model Does Well
-✅ Excellent instruction following for security requirements
-✅ Consistent, predictable responses (Google quality)
-✅ Strong code completion with security awareness
-✅ Reliable implementation of specified security controls
-✅ Clear, well-structured code generation
-✅ Native GCP integration
-### What This Model Doesn't Do
-❌ **Not a security scanner** - Use tools like Semgrep, CodeQL, or Snyk
-❌ **Not a penetration testing tool** - Cannot perform active exploitation
-❌ **Not legal/compliance advice** - Consult security professionals
-❌ **Not a replacement for security experts** - Critical systems need professional review
-❌ **Not the largest context window** - 8K tokens (vs Qwen's 128K)
-### Known Characteristics
-- **Instruction-focused:** Excels when given clear, structured requirements
-- **Consistent outputs:** Highly predictable - good for automation
-- **Google ecosystem:** Best performance when deployed on GCP
-- **Standard context:** 8K tokens sufficient for most code files
-### Appropriate Use
-✅ API generation with precise security requirements
-✅ Code completion and IDE integration
-✅ Educational platforms and training
-✅ GCP-based development workflows
-✅ Teams valuing Google engineering culture
-### Inappropriate Use
-❌ Sole security validation for production systems
-❌ Replacement for professional security audits
-❌ Active penetration testing without authorization
-❌ Very large codebase analysis (use Qwen 14B instead)
----
-## 🔬 Dataset Information
-This model was trained on **[SecureCode v2.0](https://huggingface.co/datasets/scthornton/securecode-v2)**, a production-grade security dataset with:
-- **1,209 total examples** (841 train / 175 validation / 193 test)
-- **100% incident grounding** - every example tied to real CVEs or security breaches
-- **11 vulnerability categories** - complete OWASP Top 10:2025 coverage
-- **11 programming languages** - from Python to Rust
-- **4-turn conversational structure** - mirrors real developer-AI workflows
-- **100% expert validation** - reviewed by independent security professionals
-See the [full dataset card](https://huggingface.co/datasets/scthornton/securecode-v2) and [research paper](https://perfecxion.ai/articles/securecode-v2-dataset-paper.html) for complete details.
----
-## 🏢 About perfecXion.ai
-[perfecXion.ai](https://perfecxion.ai) is dedicated to advancing AI security through research, datasets, and production-grade security tooling.
-**Connect:**
-- Website: [perfecxion.ai](https://perfecxion.ai)
-- Research: [perfecxion.ai/research](https://perfecxion.ai/research)
-- Knowledge Hub: [perfecxion.ai/knowledge](https://perfecxion.ai/knowledge)
-- GitHub: [@scthornton](https://github.com/scthornton)
-- HuggingFace: [@scthornton](https://huggingface.co/scthornton)
-- Email: scott@perfecxion.ai
----
-## 📄 License
-**Model License:** Apache 2.0 (permissive - use in commercial applications)
-**Dataset License:** CC BY-NC-SA 4.0 (non-commercial with attribution)
-### What You CAN Do
-✅ Use this model commercially in production applications
-✅ Fine-tune further for your specific use case
-✅ Deploy in enterprise environments
-✅ Integrate into commercial products
-✅ Distribute and modify the model weights
-✅ Charge for services built on this model
-### What You CANNOT Do with the Dataset
-❌ Sell or redistribute the raw SecureCode v2.0 dataset commercially
-❌ Use the dataset to train commercial models without releasing under the same license
-❌ Remove attribution or claim ownership of the dataset
-For commercial dataset licensing or custom training, contact: scott@perfecxion.ai
----
-## 📚 Citation
-If you use this model in your research or applications, please cite:
 ```bibtex
-@misc{thornton2025securecode-codegemma7b,
-  title={CodeGemma 7B - SecureCode Edition},
   author={Thornton, Scott},
-  year={2025},
   publisher={perfecXion.ai},
-  url={https://huggingface.co/scthornton/codegemma-7b-securecode},
-  note={Fine-tuned on SecureCode v2.0: https://huggingface.co/datasets/scthornton/securecode-v2}
-}
-@misc{thornton2025securecode-dataset,
-  title={SecureCode v2.0: A Production-Grade Dataset for Training Security-Aware Code Generation Models},
-  author={Thornton, Scott},
-  year={2025},
-  month={January},
-  publisher={perfecXion.ai},
-  url={https://perfecxion.ai/articles/securecode-v2-dataset-paper.html},
-  note={Dataset: https://huggingface.co/datasets/scthornton/securecode-v2}
 }
 ```
----
-## 🙏 Acknowledgments
-- **Google DeepMind & Google AI** for the excellent CodeGemma base model
-- **OWASP Foundation** for maintaining the Top 10 vulnerability taxonomy
-- **MITRE Corporation** for the CVE database and vulnerability research
-- **Security research community** for responsible disclosure practices
-- **Hugging Face** for model hosting and inference infrastructure
-- **GCP users** who validated this model in production environments
----
-## 🤝 Contributing
-Found a security issue or have suggestions for improvement?
-- 🐛 **Report issues:** [GitHub Issues](https://github.com/scthornton/securecode-models/issues)
-- 💬 **Discuss improvements:** [HuggingFace Discussions](https://huggingface.co/scthornton/codegemma-7b-securecode/discussions)
-- 📧 **Contact:** scott@perfecxion.ai
-### Community Contributions Welcome
-Especially interested in:
-- **GCP deployment examples** and Vertex AI integrations
-- **Benchmark evaluations** on security datasets
-- **Instruction-following assessments** for security tasks
-- **Production deployment case studies**
-- **Performance optimization** for GCP infrastructure
----
-## 🔗 SecureCode Model Collection
-Explore other SecureCode fine-tuned models optimized for different use cases:
-### Entry-Level Models (3-7B)
-- **[llama-3.2-3b-securecode](https://huggingface.co/scthornton/llama-3.2-3b-securecode)**
-  - **Best for:** Consumer hardware, IDE integration, education
-  - **Hardware:** 8GB RAM minimum
-  - **Unique strength:** Most accessible
-- **[deepseek-coder-6.7b-securecode](https://huggingface.co/scthornton/deepseek-coder-6.7b-securecode)**
-  - **Best for:** Security-optimized baseline
-  - **Hardware:** 16GB RAM
-  - **Unique strength:** Security-first architecture
-- **[qwen2.5-coder-7b-securecode](https://huggingface.co/scthornton/qwen2.5-coder-7b-securecode)**
-  - **Best for:** Best code understanding in 7B class
-  - **Hardware:** 16GB RAM
-  - **Unique strength:** 128K context, best-in-class
-- **[codegemma-7b-securecode](https://huggingface.co/scthornton/codegemma-7b-securecode)** ⭐ (YOU ARE HERE)
-  - **Best for:** Google ecosystem, instruction following
-  - **Hardware:** 16GB RAM
-  - **Unique strength:** Google quality, GCP integration
-### Mid-Range Models (13-15B)
-- **[codellama-13b-securecode](https://huggingface.co/scthornton/codellama-13b-securecode)**
-  - **Best for:** Enterprise trust, Meta brand
-  - **Hardware:** 24GB RAM
-  - **Unique strength:** Proven track record
-- **[qwen2.5-coder-14b-securecode](https://huggingface.co/scthornton/qwen2.5-coder-14b-securecode)**
-  - **Best for:** Advanced code analysis
-  - **Hardware:** 32GB RAM
-  - **Unique strength:** 128K context window
-- **[starcoder2-15b-securecode](https://huggingface.co/scthornton/starcoder2-15b-securecode)**
-  - **Best for:** Multi-language projects (600+ languages)
-  - **Hardware:** 32GB RAM
-  - **Unique strength:** Broadest language support
-### Enterprise-Scale Models (20B+)
-- **[granite-20b-code-securecode](https://huggingface.co/scthornton/granite-20b-code-securecode)**
-  - **Best for:** Enterprise-scale, IBM trust
-  - **Hardware:** 48GB RAM
-  - **Unique strength:** Largest model, deepest analysis
-**View Complete Collection:** [SecureCode Models](https://huggingface.co/collections/scthornton/securecode)
----
-<div align="center">
-**Built with ❤️ for secure software development**
-[perfecXion.ai](https://perfecxion.ai) | [Research](https://perfecxion.ai/research) | [Knowledge Hub](https://perfecxion.ai/knowledge) | [Contact](mailto:scott@perfecxion.ai)
----
-*Google quality. Security expertise. Production ready.*
-</div>

 ---
+license: gemma
 base_model: google/codegemma-7b-it
 tags:
+  - security
+  - cybersecurity
+  - secure-coding
+  - ai-security
+  - owasp
+  - code-generation
+  - qlora
+  - lora
+  - fine-tuned
+  - securecode
 datasets:
+  - scthornton/securecode
+library_name: peft
 pipeline_tag: text-generation
+language:
+  - code
+  - en
 ---
+# CodeGemma 7B SecureCode
 <div align="center">
+![Parameters](https://img.shields.io/badge/params-7B-blue.svg)
+![Dataset](https://img.shields.io/badge/dataset-2,185_examples-green.svg)
+![OWASP](https://img.shields.io/badge/OWASP-Top_10_2021_+_LLM_Top_10_2025-orange.svg)
+![Method](https://img.shields.io/badge/method-QLoRA_4--bit-purple.svg)
+**Security-specialized code model fine-tuned on the [SecureCode](https://huggingface.co/datasets/scthornton/securecode) dataset**
+[Dataset](https://huggingface.co/datasets/scthornton/securecode) | [Paper (arXiv:2512.18542)](https://arxiv.org/abs/2512.18542) | [Model Collection](https://huggingface.co/collections/scthornton/securecode) | [perfecXion.ai](https://perfecxion.ai)
 </div>
 ---
+## What This Model Does
+This model generates **secure code** when developers ask about building features. Instead of producing vulnerable implementations (like 45% of AI-generated code does), it:
+- Identifies the security risks in common coding patterns
+- Provides vulnerable *and* secure implementations side by side
+- Explains how attackers would exploit the vulnerability
+- Includes defense-in-depth guidance: logging, monitoring, SIEM integration, infrastructure hardening
+The model was fine-tuned on **2,185 security training examples** covering both traditional web security (OWASP Top 10 2021) and AI/ML security (OWASP LLM Top 10 2025).
+## Model Details
+| | |
+|---|---|
+| **Base Model** | [CodeGemma 7B IT](https://huggingface.co/google/codegemma-7b-it) |
+| **Parameters** | 7B |
+| **Architecture** | Gemma |
+| **Tier** | Tier 2: Mid-size Code Specialist |
+| **Method** | QLoRA (4-bit NormalFloat quantization) |
+| **LoRA Rank** | 16 (alpha=32) |
+| **Target Modules** | `q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj` (7 modules) |
+| **Training Data** | [scthornton/securecode](https://huggingface.co/datasets/scthornton/securecode) (2,185 examples) |
+| **Hardware** | NVIDIA A100 40GB |
+Google's code-specialized Gemma variant. Strong instruction following with efficient architecture.
+## Quick Start
 ```python
 from peft import PeftModel
 from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
+import torch
+# Load with 4-bit quantization (matches training)
 bnb_config = BitsAndBytesConfig(
     load_in_4bit=True,
     bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.bfloat16,
 )
+base_model = AutoModelForCausalLM.from_pretrained(
     "google/codegemma-7b-it",
     quantization_config=bnb_config,
     device_map="auto",
 )
+tokenizer = AutoTokenizer.from_pretrained("scthornton/codegemma-7b-securecode")
+model = PeftModel.from_pretrained(base_model, "scthornton/codegemma-7b-securecode")
+# Ask a security-relevant coding question
+messages = [
+    {"role": "user", "content": "How do I implement JWT authentication with refresh tokens in Python?"}
+]
+inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
+outputs = model.generate(inputs, max_new_tokens=2048, temperature=0.7)
+print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
+## Training Details
+### Dataset
+Trained on the full **[SecureCode](https://huggingface.co/datasets/scthornton/securecode)** unified dataset:
+- **2,185 total examples** (1,435 web security + 750 AI/ML security)
+- **20 vulnerability categories** across OWASP Top 10 2021 and OWASP LLM Top 10 2025
+- **12+ programming languages** and **49+ frameworks**
+- **4-turn conversational structure**: feature request, vulnerable/secure implementations, advanced probing, operational guidance
+- **100% incident grounding**: every example tied to real CVEs, vendor advisories, or published attack research
+### Hyperparameters
+| Parameter | Value |
+|-----------|-------|
+| LoRA rank | 16 |
+| LoRA alpha | 32 |
+| LoRA dropout | 0.05 |
+| Target modules | 7 linear layers |
+| Quantization | 4-bit NormalFloat (NF4) |
+| Learning rate | 2e-4 |
+| LR scheduler | Cosine with 100-step warmup |
+| Epochs | 3 |
+| Per-device batch size | 2 |
+| Gradient accumulation | 8x |
+| Effective batch size | 16 |
+| Max sequence length | 4096 tokens |
+| Optimizer | paged_adamw_8bit |
+| Precision | bf16 |
+**Notes:** Requires `trust_remote_code=True`. Extended 4096-token context for full security conversations.
+## Security Coverage
+### Web Security (1,435 examples)
+OWASP Top 10 2021: Broken Access Control, Cryptographic Failures, Injection, Insecure Design, Security Misconfiguration, Vulnerable Components, Authentication Failures, Software Integrity Failures, Logging/Monitoring Failures, SSRF.
+Languages: Python, JavaScript, Java, Go, PHP, C#, TypeScript, Ruby, Rust, Kotlin, YAML.
+### AI/ML Security (750 examples)
+OWASP LLM Top 10 2025: Prompt Injection, Sensitive Information Disclosure, Supply Chain Vulnerabilities, Data/Model Poisoning, Improper Output Handling, Excessive Agency, System Prompt Leakage, Vector/Embedding Weaknesses, Misinformation, Unbounded Consumption.
+Frameworks: LangChain, OpenAI, Anthropic, HuggingFace, LlamaIndex, ChromaDB, Pinecone, FastAPI, Flask, vLLM, CrewAI, and 30+ more.
+## SecureCode Model Collection
+This model is part of the **SecureCode** collection of 8 security-specialized models:
+| Model | Base | Size | Tier | HuggingFace |
+|-------|------|------|------|-------------|
+| Llama 3.2 SecureCode | meta-llama/Llama-3.2-3B-Instruct | 3B | Accessible | [`llama-3.2-3b-securecode`](https://huggingface.co/scthornton/llama-3.2-3b-securecode) |
+| Qwen2.5 Coder SecureCode | Qwen/Qwen2.5-Coder-7B-Instruct | 7B | Mid-size | [`qwen2.5-coder-7b-securecode`](https://huggingface.co/scthornton/qwen2.5-coder-7b-securecode) |
+| DeepSeek Coder SecureCode | deepseek-ai/deepseek-coder-6.7b-instruct | 6.7B | Mid-size | [`deepseek-coder-6.7b-securecode`](https://huggingface.co/scthornton/deepseek-coder-6.7b-securecode) |
+| CodeGemma SecureCode | google/codegemma-7b-it | 7B | Mid-size | [`codegemma-7b-securecode`](https://huggingface.co/scthornton/codegemma-7b-securecode) |
+| CodeLlama SecureCode | codellama/CodeLlama-13b-Instruct-hf | 13B | Large | [`codellama-13b-securecode`](https://huggingface.co/scthornton/codellama-13b-securecode) |
+| Qwen2.5 Coder 14B SecureCode | Qwen/Qwen2.5-Coder-14B-Instruct | 14B | Large | [`qwen2.5-coder-14b-securecode`](https://huggingface.co/scthornton/qwen2.5-coder-14b-securecode) |
+| StarCoder2 SecureCode | bigcode/starcoder2-15b-instruct-v0.1 | 15B | Large | [`starcoder2-15b-securecode`](https://huggingface.co/scthornton/starcoder2-15b-securecode) |
+| Granite 20B Code SecureCode | ibm-granite/granite-20b-code-instruct-8k | 20B | XL | [`granite-20b-code-securecode`](https://huggingface.co/scthornton/granite-20b-code-securecode) |
+Choose based on your deployment constraints: **3B** for edge/mobile, **7B** for general use, **13B-15B** for deeper reasoning, **20B** for maximum capability.
+## SecureCode Dataset Family
+| Dataset | Examples | Focus | Link |
+|---------|----------|-------|------|
+| **SecureCode** | 2,185 | Unified (web + AI/ML) | [scthornton/securecode](https://huggingface.co/datasets/scthornton/securecode) |
+| SecureCode Web | 1,435 | Web security (OWASP Top 10 2021) | [scthornton/securecode-web](https://huggingface.co/datasets/scthornton/securecode-web) |
+| SecureCode AI/ML | 750 | AI/ML security (OWASP LLM Top 10 2025) | [scthornton/securecode-aiml](https://huggingface.co/datasets/scthornton/securecode-aiml) |
+## Intended Use
+**Use this model for:**
+- Training AI coding assistants to write secure code
+- Security education and training
+- Vulnerability research and secure code review
+- Building security-aware development tools
+**Do not use this model for:**
+- Offensive exploitation or automated attack generation
+- Circumventing security controls
+- Any activity that violates the base model's license
+## Citation
 ```bibtex
+@misc{thornton2026securecode,
+  title={SecureCode: A Production-Grade Multi-Turn Dataset for Training Security-Aware Code Generation Models},
   author={Thornton, Scott},
+  year={2026},
   publisher={perfecXion.ai},
+  url={https://huggingface.co/datasets/scthornton/securecode},
+  note={arXiv:2512.18542}
 }
 ```
+## Links
+- **Dataset**: [scthornton/securecode](https://huggingface.co/datasets/scthornton/securecode)
+- **Research Paper**: [arXiv:2512.18542](https://arxiv.org/abs/2512.18542)
+- **Model Collection**: [huggingface.co/collections/scthornton/securecode](https://huggingface.co/collections/scthornton/securecode)
+- **Author**: [perfecXion.ai](https://perfecxion.ai)
+## License
+This model is released under the **gemma** license (inherited from the base model). The training dataset ([SecureCode](https://huggingface.co/datasets/scthornton/securecode)) is licensed under **CC BY-NC-SA 4.0**.