mdk615661/it-helpdesk-merged-v4
The mdk615661/it-helpdesk-merged-v4 is a 7 billion parameter causal language model fine-tuned for IT helpdesk ticket classification. Developed by mdk615661, this model specializes in normalizing and categorizing IT support requests into predefined categories and subcategories, while also generating insights and recommendations. It was fine-tuned using QLoRA on 2,000 Qwen-generated IT helpdesk records, achieving a training loss of 0.187. This model is optimized for corporate IT support teams to automate ticket processing and improve efficiency.
Loading preview...
Overview
mdk615661/it-helpdesk-merged-v4 is a 7 billion parameter language model specifically fine-tuned for IT helpdesk ticket classification. Building upon its predecessor, v3, this version incorporates an additional 2,000 Qwen-generated IT helpdesk records during its QLoRA fine-tuning process, achieving a final training loss of 0.187 over 3 epochs. The model is designed to process raw IT support requests and provide structured outputs including category, subcategory, normalized text, priority, and actionable insights.
Key Capabilities
- Automated Ticket Classification: Accurately assigns IT helpdesk tickets to predefined categories and subcategories (e.g., Hardware - Laptop, Software - Password Reset, Incident - Network Outage).
- Insight Generation: Provides contextual insights and recommendations for each classified ticket, such as "Hardware failure preventing user from working" and "Raise repair request with IT hardware team."
- Normalization: Standardizes input text for consistent processing.
- Priority Assignment: Suggests an appropriate priority level for the ticket.
Use Cases
This model is ideal for corporate IT support environments looking to:
- Streamline Ticket Management: Automate the initial triage and routing of incoming helpdesk tickets.
- Improve Efficiency: Reduce manual effort in categorizing and understanding support requests.
- Enhance Reporting: Provide structured data for better analysis of IT support trends and common issues.
Technical Details
The model was fine-tuned using QLoRA (r=16, alpha=32) with a learning rate of 2e-4 and a cosine scheduler. It operates with fp16 precision and has a context length of 4096 tokens.