from typing import Dict, List, Any
from PIL import Image
import torch
import base64
from io import BytesIO
from transformers import AutoProcessor, BlipForConditionalGeneration

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

class EndpointHandler():
    def __init__(self, path=""):
        self.processor = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
        self.model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large").to(device)

    def __call__(self, data: Any) -> List[float]:
        inputs = data.pop("inputs", data)

        image = Image.open(BytesIO(base64.b64decode(inputs['image'])))
        inputs = self.processor(image, inputs['text'], return_tensors="pt").to(device)
        outputs = self.model.generate(**inputs)
        
        return self.processor.decode(outputs[0], skip_special_tokens=True)