Finetuned Mistral-7B Ticket Summarizer

This model is a fine-tuned version of Mistral-7B Instruct v3, designed to clean and concisely summarize structured support ticket entries. Each output follows a strict, two-part format:

1. Issue Summary: A brief overview of the user-reported issue
   
2. Resolution: A concise description of the support team’s response

Key Features

• Consistent formatting for downstream automation (e.g., clustering, FAQ generation)
  
• Low hallucination rate through role-based prompt tuning
  
• Lightweight deployment enabled by 8-bit quantization (Q8_0)
  
• Instruction-following fine-tuning via LoRA adapters
  

Prompt Format

The model was fine-tuned using a role-based instruction prompt designed to enforce strict summarization structure:

Deployment Options

• Ollama (Local Deployment) Run the quantized model locally
• GenAI Studio API (Hosted Access) This model is accessible via Purdue’s GenAI Studio for programmatic summarization.
• HuggingFace Model weights available at cjoslin22/hpc-ticket/summarizer/mistral7b (private repository)

Intended Use

This model is intended for structured summarization of internal or enterprise-level support tickets, where brevity, consistency, and format control are essential.

It is not intended for open-ended conversation, creative generation, or use with unfiltered or public-facing content

Author

Christina Joslin RCAC Intern, June 2025