48 lines
2.1 KiB
Markdown
48 lines
2.1 KiB
Markdown
---
|
|
language:
|
|
- en
|
|
- ur
|
|
license: apache-2.0
|
|
tags:
|
|
- medical
|
|
- clinical
|
|
- icd-10
|
|
- billing
|
|
- mlx
|
|
pipeline_tag: text-generation
|
|
base_model: Qwen/Qwen2.5-0.5B-Instruct
|
|
widget:
|
|
- text: "<system>You are the Cognitapp Global ICD-10 Assistant. Extract the primary ICD-10 code.</system> <user>Patient with severe high fever, joint pain, and suspected Dengue from Lahore.</user> <assistant>"
|
|
example_title: "Dengue Case (Regional)"
|
|
- text: "<system>You are the Cognitapp Global ICD-10 Assistant. Extract the primary ICD-10 code.</system> <user>65yo male smoker with chronic cough and SOB. Spirometry shows FEV1/FVC < 0.70.</user> <assistant>"
|
|
example_title: "COPD Case (Global)"
|
|
---
|
|
|
|
# Cognitapp-Med-Nano-v1
|
|
|
|
**Cognitapp-Med-Nano-v1** is a specialized, lightweight medical large language model (LLM) developed by Cognitapp Labs. It is fine-tuned from the Qwen2.5-0.5B architecture to excel at **ICD-10-CM Medical Billing and Clinical Extraction**.
|
|
|
|
## Key Features
|
|
- **Global & Regional Awareness:** Optimized for both international clinical standards.
|
|
- **Efficiency:** 0.5B parameters, designed for 100% offline use on mobile and desktop devices via MLX or llama.cpp.
|
|
- **Precision:** Trained using prompt-masking to prioritize alphanumeric code accuracy over conversational filler.
|
|
|
|
## How to use with MLX
|
|
```python
|
|
from mlx_lm import load, generate
|
|
|
|
model, tokenizer = load("Cognitapp/Cognitapp-Med-Nano-v1")
|
|
prompt = "<system>You are the Cognitapp Global ICD-10 Assistant. Extract the primary ICD-10 code.</system> <user>Patient has 103F fever, body aches, and positive NS1 for Dengue.</user> <assistant>"
|
|
response = generate(model, tokenizer, prompt=prompt, max_tokens=10)
|
|
print(response)
|
|
```
|
|
|
|
## Intended Use
|
|
This model is a supportive tool for medical professionals and billers. It is NOT a diagnostic tool.
|
|
|
|
## Training Data
|
|
Fine-tuned on a balanced dataset of 1,200+ global and regional clinical scenarios including pediatrics, geriatrics, and infectious diseases.
|
|
|
|
## Disclaimer
|
|
All outputs must be verified by a licensed healthcare professional. Cognitapp Labs is not responsible for any clinical or billing errors.
|