deepmodel/dm-llm-hc-sm

Model Details

Model Description

The dm-llm-hc-sm model has been fine-tuned on Llama-3.2-3B-Instruct model, and has been further adapted to the domain-specific task through fine-tuning.

Model Vesion:

• v1.0

Developed by

• deepmodel

Usecase:

*The LLM model evaluates Prior Authorization (PA) forms by analyzing the provided medical history and guidelines to generate a decision (Accept, Reject, Missing Information) and a justification. It ensures compliance with medical policies and identifies gaps or issues in submitted documentation.

Input:

• Prior Authorization (PA) form
  • Prior Authorization (PA) form is a document used to obtain approval from an insurance company before a healthcare provider administers a treatment, prescription, or procedure.
• Medical History:
  • Medical history refers to a comprehensive record of a patient’s past and present health conditions, treatments, medications, surgeries, allergies, and family health history.
• Medical Guidline Content:
  • A medical guideline is a structured set of recommendations based on clinical evidence, aimed at optimizing patient care.

Outputs

• Decision (Accept, Reject, Mission Information)
• Justification.

Fine-tuning Techniques Used

1. Low-Rank Adaptation (LoRA): The model uses LoRA (Low-Rank Adaptation), a technique that adapts pre-trained models efficiently by adding low-rank matrices to certain layers.
2. Quantization: The model has been quantized using 4-bit quantization from Bits and Bytes (bnb) library to reduce memory usage and computational cost while maintaining performance.
3. Task-Specific Training: Fine-tuning involved training on a synthetic dataset including patient's prior authorization forms , medical histories, and medical guideline contents.