handler.py

from typing import  Dict, List, Any
from PIL import Image
import torch
from torch import autocast
from diffusers import StableDiffusionPipeline
import base64
from io import BytesIO
from transformers.utils import logging

logging.set_verbosity_info()
logger = logging.get_logger("transformers")
logger.info("INFO")
logger.warning("WARN")


# 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=""):
        self.path = path
        # load the optimized model
        model_id = "stabilityai/stable-diffusion-x4-upscaler"
        self.pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
        self.pipe = self.pipe.to(device)


    def __call__(self, data) -> List[Dict[str, Any]]:
        """
        Args:
            image (:obj:`string`)
        Return:
            A :obj:`dict`:. base64 encoded image
        """
        logger.info('data received %s', data)
        inputs = data.get("inputs")
        logger.info('inputs received %s', inputs)

        decoded_image = Image.open(BytesIO(base64.b64decode(inputs['image'])))
        with autocast(device.type):
            upscaled_image = self.pipe(prompt="", image = decoded_image).images[0]

        buffered = BytesIO()
        upscaled_image.save(buffered, format="JPEG")
        img_str = base64.b64encode(buffered.getvalue())

        # postprocess the prediction
        return {"image": img_str}