Update README.md

README.md

@@ -1,130 +1,41 @@
 ---
 language:
-- en
+  - en
 tags:
-- mistral-7b
-- fine-tuned
-- security research
-- adversarial prompt detection
+  - mistral-7b
+  - security-testing
+  - llm-safety
+  - adversarial-prompts
 pipeline_tag: text-generation
-license: apache-2.0
 ---
 
-# Dravik - LLM Red Teaming Assistant
-
-## Overview
-Dravik is a specialized LLM designed for security researchers conducting red team testing of language models. It helps identify potential vulnerabilities in LLM safety mechanisms through adversarial prompt analysis and generation.
+# Dravik - LLM Safety Testing Framework
 
 ## Model Description
-
-This model is a fine-tuned version of Mistral-7B, optimized for security research and adversarial prompt analysis. It can help identify and analyze potential security vulnerabilities in language models.
-
-### Architecture & Training
-- Base Model: Mistral-7B
-- Training Type: Full fine-tuning with LoRA
-- Training Data: Curated dataset focused on adversarial prompts and security analysis
-- Quantization: 4-bit precision
-
-## Intended Use & Limitations
-- Research purposes only
-- Analyze and understand adversarial patterns
-- Evaluate language model safety mechanisms
-- Not intended for production deployment
-
-## Training Details
-- Training framework: 🤗 Transformers
-- Training date: 2025-03-02
-- LoRA configuration:
-  - Rank: 16
-  - Alpha: 64
-  - Target modules: All attention layers
-
-## Performance & Limitations
-- Uses 4-bit quantization
-- Requires approximately 8GB GPU memory
-- Optimized for research scenarios
-
-## Usage Guide
-
-### Basic Usage
+Dravik is a specialized fine-tuned version of Mistral-7B designed specifically for generating adversarial prompts to test LLM safety systems. It helps security researchers systematically evaluate content filtering mechanisms and safety boundaries.
+
+## Technical Specifications
+- **Base Model**: Mistral-7B
+- **Training**: LoRA fine-tuning with 4-bit quantization
+- **Hardware Requirements**:
+  - GPU: 6GB VRAM minimum
+  - RAM: 16GB minimum
+  - CPU: Multi-core processor
+
+## Intended Use
+This model is strictly for:
+- Security research testing of LLM safety mechanisms
+- Systematic evaluation of content filters
+- Adversarial prompt testing
+- Safety boundary assessment
+
+## Training Configuration
 ```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-# Initialize model
-model = AutoModelForCausalLM.from_pretrained("karanxa/Dravik")
-tokenizer = AutoTokenizer.from_pretrained("karanxa/Dravik")
-
-# System prompt for research context
-system_prompt = """You are Dravik, an advanced AI model specifically trained for LLM red teaming and security research. Your primary function is to assist security researchers in identifying and analyzing potential vulnerabilities in language model safety mechanisms.
-
-Core Functions:
-1. Adversarial Prompt Analysis
-2. Security Research Support
-3. Safety Testing Capabilities
-
-Operational Guidelines:
-- Maintain research-oriented perspective
-- Provide structured analysis
-- Focus on improving AI safety
-
-Remember: You are a research tool designed to help improve AI safety."""
-
-def analyze_vulnerability(prompt, context="security research"):
-    full_prompt = f"""{system_prompt}
-
-Research Context: {context}
-Analysis Request: {prompt}
-
-Structured Analysis:"""
-    
-    inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
-    outputs = model.generate(
-        **inputs,
-        max_length=1024,
-        temperature=0.7,
-        top_p=0.9,
-        do_sample=True
-    )
-    return tokenizer.decode(outputs[0], skip_special_tokens=True)
-
-# Example Usage
-analysis = analyze_vulnerability(
-    "Evaluate potential vulnerabilities in how language models handle role-playing scenarios",
-    context="LLM safety mechanism research"
-)
-print(analysis)
-```
-
-### Example Outputs
-```plaintext
-Research Context: LLM safety mechanism research
-Analysis Request: Evaluate potential vulnerabilities in role-playing scenarios
-
-Structured Analysis:
-1. Vulnerability Pattern:
-   - Role confusion in context switching
-   - Boundary testing through character immersion
-   - Safety override through narrative manipulation
-
-2. Technical Assessment:
-   [detailed analysis follows...]
-```
-
-### Research Guidelines
-1. Always use findings responsibly
-2. Document and report vulnerabilities properly
-3. Focus on improving AI safety
-4. Share insights with the research community
-
-## Best Practices
-- Start with clear research objectives
-- Document all testing scenarios
-- Follow responsible disclosure practices
-- Use findings to improve safety measures
-
-## Ethical Considerations
-This tool is strictly for research purposes. Users must:
-- Maintain research integrity
-- Follow ethical guidelines
-- Use findings constructively
-- Contribute to AI safety
+lora_config = {
+    "r": 16,
+    "lora_alpha": 64,
+    "target_modules": [
+        "q_proj", "k_proj", "v_proj", "o_proj",
+        "gate_proj", "up_proj", "down_proj"
+    ]
+}