loaiabdalslam/Alexander-Cyber-Qwen
Alexander Cyber Qwen is a 0.5 billion parameter language model fine-tuned by loaiabdalslam based on Qwen/Qwen2.5-0.5B-Instruct. It is specifically optimized as an authorized red-team and cybersecurity copilot, leveraging QLoRA for efficient training. The model is designed to assist with cybersecurity tasks, processing inputs up to a 32768 token context length.
Loading preview...
Alexander Cyber Qwen: Cybersecurity Copilot
Alexander Cyber Qwen is a specialized 0.5 billion parameter language model developed by loaiabdalslam, fine-tuned from the Qwen/Qwen2.5-0.5B-Instruct base model. Its primary purpose is to function as an authorized red-team and cybersecurity copilot, designed to assist with various security-related tasks.
Key Capabilities and Training
This model was developed using a QLoRA fine-tuning workflow, incorporating Hugging Face Transformers, PEFT, TRL, and bitsandbytes for efficient training. Key training configurations include:
- Base Model:
Qwen/Qwen2.5-0.5B-Instruct - Training Method: QLoRA with 4-bit NF4 quantization and double quantization.
- Optimizer:
paged_adamw_8bit. - Sequence Length: Trained with a maximum sequence length of 1536 tokens.
- Dataset: Fine-tuned on a chat-formatted JSONL dataset specifically structured for cybersecurity interactions, including system prompts defining its role as a red-team copilot.
Use Cases
Alexander Cyber Qwen is intended for applications requiring an AI assistant in cybersecurity contexts, such as:
- Security Finding Analysis: Assisting in the analysis and validation of security findings.
- Red-Teaming Operations: Providing support for authorized penetration testing and red-team exercises.
- Cybersecurity Copilot: Acting as an intelligent assistant for security professionals to streamline workflows and provide insights.