handler.py

import requests
from typing import Dict, Any
from PIL import Image
import torch
import base64
from io import BytesIO
from transformers import BlipForConditionalGeneration, BlipProcessor

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

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

    def __call__(self, data: Dict[str, Any]) -> Dict[str, Any]:
        input_data = data.get("inputs", {})
        encoded_images = input_data.get("images")
        
        if not encoded_images:
            return {"captions": [], "error": "No images provided"} 

        texts = input_data.get("texts", ["a photography of"] * len(encoded_images)) 

        try:
            raw_images = [Image.open(BytesIO(base64.b64decode(img))).convert("RGB") for img in encoded_images]
            processed_inputs = [
                self.processor(image, text, return_tensors="pt") for image, text in zip(raw_images, texts)
            ]
            processed_inputs = {
                "pixel_values": torch.cat([inp["pixel_values"] for inp in processed_inputs], dim=0).to(device),
                "input_ids": torch.cat([inp["input_ids"] for inp in processed_inputs], dim=0).to(device),
                "attention_mask": torch.cat([inp["attention_mask"] for inp in processed_inputs], dim=0).to(device)
            }

            with torch.no_grad():
                out = self.model.generate(**processed_inputs)

            captions = self.processor.batch_decode(out, skip_special_tokens=True)
            return {"captions": captions}
        except Exception as e:
            print(f"Error during processing: {str(e)}")
            return {"captions": [], "error": str(e)}