metadata
license: apache-2.0
language:
- en
base_model:
- unsloth/gemma-4-e4b-it-unsloth-bnb-4bit
tags:
- transformers
- safetensors
- gemma4
- image-text-to-text
- text-generation-inference
- unsloth
- conversational
- icd-coding
- clinical-nlp
- medical-ai
pipeline_tag: image-text-to-text
🏥 gemma4-e4b-icd-coding
A fine-tuned Gemma 4 E4B model for automatic ICD code prediction from clinical notes. Given a free-text clinical note, the model outputs the relevant ICD diagnosis codes — streamlining medical billing, documentation, and clinical analytics workflows.
📌 Model Overview
| Property | Details |
|---|---|
| Base Model | unsloth/gemma-4-e4b-it-unsloth-bnb-4bit |
| Fine-tuned by | nikhil061307 |
| Task | Clinical Note → ICD Code Prediction |
| Language | English |
| License | Apache 2.0 |
| Training Framework | Unsloth + HuggingFace TRL |
🚀 What It Does
Given a clinical note like:
"Patient presents with persistent cough, fever, and bilateral infiltrates on chest X-ray. Diagnosed with community-acquired pneumonia."
The model outputs the appropriate ICD-10 code(s), e.g.:
J18.9 - Pneumonia, unspecified organism
💻 Usage
Installation
pip install unsloth transformers torch
Inference
from transformers import AutoTokenizer, AutoModelForImageTextToText
import torch
model_id = "nikhil061307/gemma4-e4b-icd-coding"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(
model_id,
device_map="auto",
torch_dtype=torch.bfloat16,
)
clinical_note = """
Patient is a 65-year-old male with a history of type 2 diabetes presenting
with polyuria, polydipsia, and HbA1c of 9.2%. Blood glucose fasting at 210 mg/dL.
"""
messages = [
{
"role": "user",
"content": f"Predict the ICD-10 codes for the following clinical note:\n\n{clinical_note}"
}
]
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
with torch.no_grad():
outputs = model.generate(
**inputs,
max_new_tokens=256,
temperature=0.1,
do_sample=True,
)
response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True)
print(response)
🧪 Example Input / Output
Input (Clinical Note):
A 52-year-old woman presents with sharp chest pain radiating to the left arm,
diaphoresis, and shortness of breath. ECG shows ST elevation in leads II, III, aVF.
Troponin elevated. Impression: Acute inferior STEMI.
Output (ICD Codes):
I21.19 - ST elevation (STEMI) myocardial infarction involving other coronary artery
⚙️ Training Details
- Base model:
unsloth/gemma-4-e4b-it-unsloth-bnb-4bit(4-bit quantized) - Training speedup: 2x faster training with Unsloth
- Library: HuggingFace TRL (SFTTrainer)
- Quantization: BnB 4-bit (inference efficient)
⚠️ Limitations & Disclaimer
- This model is intended for research and assistive purposes only.
- It is not a substitute for professional medical coding by certified coders (CPC/CCS).
- Always verify predicted ICD codes with qualified clinical staff before use in billing or official documentation.
- Model performance may vary across specialties, note styles, and rare diagnosis categories.
📄 License
This model is released under the Apache 2.0 license. See LICENSE for details.
🙏 Acknowledgements
- Unsloth AI — for the blazing fast fine-tuning framework
- Google DeepMind — for the Gemma model family
- HuggingFace TRL — for the SFT training utilities
Made with ❤️ using Unsloth