from typing import Dict, Any
from PIL import Image    
import requests
import torch
import numpy as np
from transformers import AutoProcessor, LlavaForConditionalGeneration

class EndpointHandler():
    def __init__(self, path=""):
        model_id = path
        self.model = LlavaForConditionalGeneration.from_pretrained(
            model_id, 
            torch_dtype=torch.float16, 
            low_cpu_mem_usage=True, 
        ).to(0)
        self.processor = AutoProcessor.from_pretrained(model_id)

    def __call__(self, data: Dict[str, Any]):
        parameters = data.pop("inputs", data)
        if parameters is not None:
            url = "http://images.cocodataset.org/val2017/000000039769.jpg"
            prompt = "USER: <image>\nWhat are these?\nASSISTANT:"
            raw_image = Image.open(requests.get(url, stream=True).raw)
            inputs = self.processor(prompt, raw_image, return_tensors='pt').to(0, torch.float16)
            output = self.model.generate(**inputs, max_new_tokens=200, do_sample=False)
            # Convert Tensor to NumPy array or list before returning
            output = output.cpu().numpy().tolist()
        return output