app.py

import os
import re
import gradio as gr
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

hf_token = os.environ.get("HF_TOKEN")

MODEL_ID = "StudioIlios/icd10-model"
BASE_MODEL = "unsloth/Llama-3.2-3B-Instruct-bnb-4bit"

print("Loading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(
MODEL_ID,
token=hf_token
)

if tokenizer.pad_token_id is None:
    tokenizer.pad_token_id = tokenizer.eos_token_id

print("Loading base model...")
base_model = AutoModelForCausalLM.from_pretrained(
BASE_MODEL,
token=hf_token,
device_map="auto"
)

print("Loading LoRA adapter...")
model = PeftModel.from_pretrained(
base_model,
MODEL_ID,
token=hf_token
)

model.eval()

print("Model loaded successfully!")

def predict(clinical_text):

    prompt = f"""
Patient has {clinical_text}.
What is the ICD10 code?
"""

    inputs = tokenizer(
        prompt,
        return_tensors="pt"
    )

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=20,
            do_sample=False,
            pad_token_id=tokenizer.eos_token_id
        )

    full_output = tokenizer.decode(
        outputs[0],
        skip_special_tokens=True
    )

    print("INPUT:", clinical_text)
    print("OUTPUT:", full_output)

    icd_match = re.search(
        r'\b[A-Z][0-9][A-Z0-9](?:\.[A-Z0-9]{1,4})?\b',
        full_output
    )

    if icd_match:
        return icd_match.group(0)

    return full_output

demo = gr.Interface(
fn=predict,
inputs=gr.Textbox(
label="Clinical Text",
lines=4
),
outputs=gr.Textbox(
label="Predicted ICD-10 Code"
),
title="ICD-10 Predictor"
)

demo.launch()