dvr76/ticket-triage-qwen3
The dvr76/ticket-triage-qwen3 model is a fine-tuned Qwen3-2B language model developed by dvr76, specifically optimized for extracting structured maintenance information from tenant ticket text. This model leverages QLoRA fine-tuning with 4-bit NF4 quantization to efficiently process and categorize maintenance requests. Its primary strength lies in transforming unstructured text into a predefined JSON schema, making it ideal for automated property maintenance triage systems.
Loading preview...
Model Overview
dvr76/ticket-triage-qwen3 is a specialized language model built upon the Qwen3-2B architecture, fine-tuned to automate the processing of property maintenance requests. Its core function is to parse natural language descriptions of issues from tenant tickets and convert them into a structured JSON format, facilitating efficient triage and vendor assignment.
Key Capabilities
- Structured Information Extraction: Converts free-form text into a predefined JSON schema, identifying maintenance-related details.
- Maintenance Request Triage: Designed to determine if a request is indeed a maintenance issue and extract relevant attributes like issue category, sub-category, location, and urgency.
- Vendor Type Identification: Can suggest the appropriate vendor type required for a specific maintenance task.
- Entry Requirement Detection: Identifies whether entry is required for the maintenance work.
Training Details
The model was fine-tuned using the QLoRA method with 4-bit NF4 quantization on the base Qwen/Qwen3-2B model. Training was conducted for 3 epochs with a learning rate of 2e-4, utilizing Unsloth and TRL SFTTrainer.
Good For
- Automating initial processing of tenant maintenance tickets.
- Integrating into property management systems for efficient issue categorization.
- Reducing manual effort in triaging and routing maintenance requests.
- Applications requiring structured data extraction from short, descriptive text inputs related to property issues.