ligaments-dev/Qwen2.5-1.5B-Instruct-itr-finetuned
The ligaments-dev/Qwen2.5-1.5B-Instruct-itr-finetuned model, developed by Ligaments AI, is a 1.5 billion parameter instruction-tuned causal language model based on the Qwen2.5 architecture. It is specifically fine-tuned for extracting structured JSON data from Indian Income Tax Return (ITR) documents, including ITR-1, ITR-2, ITR-3, and ITR-4. This model excels at converting ITR document text into valid JSON with high accuracy for financial data points, making it suitable for specialized financial processing tasks.
Loading preview...
Model Overview
ligaments-dev/Qwen2.5-1.5B-Instruct-itr-finetuned is a specialized 1.5 billion parameter language model developed by Ligaments AI. It is a fine-tuned version of the Qwen/Qwen2.5-1.5B-Instruct base model, specifically optimized for a unique task: structured JSON extraction from Indian Income Tax Return (ITR) documents.
Key Capabilities
This model is designed to accurately parse and extract financial data from various ITR forms (ITR-1, ITR-2, ITR-3, ITR-4) and output it in a structured JSON format. Its fine-tuning process, utilizing LoRA with MLX-LM, has resulted in strong performance metrics:
- High JSON Validity: Achieves a 98.0% pass rate for generating valid JSON.
- Accurate Data Extraction: Demonstrates 98.0% accuracy for form type matching, numeric sum correctness, boolean Y/N fields, and date formatting (YYYY-MM-DD).
- Specialized for ITR: Tailored to handle the specific nuances of Indian tax documents.
Intended Use Cases
This model is particularly well-suited for:
- MSME Lending Workflows: Automating the extraction of financial data from ITR documents to streamline lending processes.
- Credit Risk Assessment: Enhancing automated credit risk assessment pipelines by providing structured financial inputs.
It is important to note that this model is not intended for general-purpose tax advice or legal decisions, but rather for data extraction within specific financial automation contexts.