main2.py

import fastapi
from fastapi import FastAPI, File, UploadFile
import cv2
import numpy as np
from PIL import Image
from io import BytesIO
import keras
import uvicorn
import os
# from pydantic import BaseModel
app = FastAPI()

import requests

API_URL = "https://api-inference.huggingface.co/models/google/gemma-7b"
headers = {"Authorization": "Bearer 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"}

def query(payload):
	response = requests.post(API_URL, headers=headers, json=payload)
	return response.json()



model = keras.models.load_model("Ai714")


def read_imagefile(file) -> Image.Image:
    image = Image.open(BytesIO(file))
    return image

def preprocessing(image: Image.Image):
    # Convert PIL image to NumPy array
    image_np = np.array(image)

    # Resize the image
    resized_image = cv2.resize(image_np, (420, 420))

    # Add a channel dimension to match model input shape
    resized_image = np.expand_dims(resized_image, axis=0)

    return resized_image

@app.get('/')
def index():
    return{'message':'hello world'}

@app.post("/predict/")
async def predict_image(name: str,file: UploadFile = File(...)):
    # Load the model


    # Read the uploaded image file
    image = read_imagefile(await file.read())

    # Preprocess the image
    image = preprocessing(image)

    # Maprediction
    predictione  = model.predict(image)
    max_idx=predictione.argmax()
    x=""
    if(max_idx==0):
        x="first degree burn"
    elif(max_idx==1):
        x="second degree burn"
    else:
        x="third degree burn"

	
    output = query({
	    "inputs": "what are remedies for "+x+". If a paitent alredy have "+name+" problem",
    })
    output= output[0]['generated_text']
    output = output.replace('\n', ' ')

# Remove extra spaces
    output = ' '.join(output.split())
    
    # Return prediction as response
    return {"prediction": max_idx.tolist(),
            "llm_op":output}




    # Get the port number from the environment variable if available
if __name__ == '__main__':
    uvicorn.run(app, host='127.0.0.1', port=8000)