AmareshHebbar/icd10-coder-qwen25-7b-merged
AmareshHebbar/icd10-coder-qwen25-7b-merged is a 7.6 billion parameter Qwen2.5-based decoder-only transformer model fine-tuned for medical classification. Developed by AmareshHebbar, it specializes in assigning WHO-standard ICD-10 codes, explaining classification logic, and estimating WHO-level insurance coverage from clinical descriptions. This model is optimized for healthcare applications requiring precise medical coding and preliminary insurance intelligence.
Loading preview...
ICD-10 Medical Coder — Qwen2.5-7B
This model, developed by AmareshHebbar, is the initial phase of the AxisMapper initiative, aiming to create an AI-native insurance intelligence layer for healthcare. It is a fine-tuned Qwen2.5-7B-Instruct model, specifically designed to address the opacity in medical insurance coverage decisions by standardizing medical classification.
Key Capabilities
- ICD-10 Code Assignment: Accurately assigns WHO-standard ICD-10 codes (primary, secondary, procedure) based on clinical descriptions.
- Classification Logic Explanation: Provides explanations for WHO classification logic, detailing why specific codes are assigned and their categories.
- WHO-level Coverage Estimation: Estimates insurance coverage based on WHO standards, including reimbursement brackets, admission duration, and procedure eligibility.
- Restriction Flagging: Identifies and flags critical restrictions such as minimum admission days, co-morbidity requirements, and pre-authorization triggers.
- Multi-condition Support: Handles complex scenarios involving comorbidities, complications, and dual coding.
Training and Performance
The model was fine-tuned using LoRA via Unsloth and HuggingFace TRL on an NVIDIA RTX A5000 GPU, achieving 2x faster training speed and ~60% VRAM reduction compared to standard methods. It supports a maximum sequence length of 2048 tokens and is licensed under Apache 2.0. The model covers all major ICD-10-CM chapters, from infectious diseases to injuries.
Intended Use and Limitations
This model is intended as a tool for universal code prediction and coverage logic based on WHO standards. It is not a substitute for professional medical coding or licensed insurance adjudication, as its coverage estimates are indicative and not legally binding. Future phases of AxisMapper will include scheme-specific models for more precise, policy-aware outputs for various insurance providers.