from typing import  Dict, List, Any
import torch
import random
from torch import autocast
from diffusers import DiffusionPipeline
import base64
from io import BytesIO


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

if device.type != 'cuda':
    raise ValueError("need to run on GPU")

class EndpointHandler():
    def __init__(self, path=""):
        # load the optimized model
        self.pipe = DiffusionPipeline.from_pretrained("SG161222/Realistic_Vision_V4.0")
        self.pipe = self.pipe.to(device)


    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
        """
        Args:
            data (:obj:):
                includes the input data and the parameters for the inference.
        Return:
            A :obj:`dict`:. base64 encoded image
        """
        inputs = data.pop("inputs", data)
        random_generators= [torch.Generator().manual_seed(i) for i in range(inputs["num_images_per_prompt"])]
        print(random_generators)
        
        # run inference pipeline
        with autocast(device.type):
            images = self.pipe(prompt=inputs["prompt"], negative_prompt=inputs["negative_prompt"], num_images_per_prompt=inputs["num_images_per_prompt"], guidance_scale=7.5, generator=random_generators).images
        # encode image as base 64
        buffered = BytesIO()
        base64_images = []
        for image in images:
            image.save(buffered, format="JPEG")
            img_str = base64.b64encode(buffered.getvalue())
            base64_images.append(img_str.decode())
        # postprocess the prediction
        return {"images": base64_images}