File size: 2,505 Bytes
e4d55ce
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f3b3a2f
e4d55ce
 
 
 
 
 
 
e6af97e
 
e4d55ce
 
 
 
 
 
 
 
9b2c38b
e4d55ce
 
 
 
 
 
f806422
 
 
 
 
 
 
 
 
 
 
 
e4d55ce
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
from typing import  Dict, List, Any
import torch
from PIL import Image
from io import BytesIO
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, DDIMScheduler

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

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

model_id = "stabilityai/stable-diffusion-2-1-base"

class EndpointHandler():
    def __init__(self, path=""):
        # load the optimized model
        self.textPipe = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
        self.textPipe.scheduler = DDIMScheduler.from_config(self.textPipe.scheduler.config)
        self.textPipe = self.textPipe.to(device)

        # create an img2img model
        self.imgPipe = StableDiffusionImg2ImgPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
        self.imgPipe.scheduler = DDIMScheduler.from_config(self.imgPipe.scheduler.config)
        self.imgPipe = self.imgPipe.to(device)

    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
        raise NotImplementedError()
        """
        Args:
            data (:obj:):
                includes the input data and the parameters for the inference.
        Return:
            A :obj:`dict`:. base64 encoded image
        """
        image_data = data.pop("inputs", data).decode('base64')
        init_image = Image.open(image_data).convert("RGB")
        init_image.thumbnail((512, 512))

        params = data.pop("parameters", data)

        # hyperparamters
        num_inference_steps = params.pop("num_inference_steps", 25)
        guidance_scale = params.pop("guidance_scale", 7.5)
        negative_prompt = params.pop("negative_prompt", None)
        prompt = params.pop("prompt", None)
        height = params.pop("height", None)
        width = params.pop("width", None)
        manual_seed = params.pop("manual_seed", -1)

        out = None

        generator = torch.Generator(device='cuda')
        generator.manual_seed(manual_seed)
        # run img2img pipeline
        out = self.imgPipe(prompt,
                    image=init_image,
                    num_inference_steps=num_inference_steps,
                    guidance_scale=guidance_scale,
                    num_images_per_prompt=1,
                    negative_prompt=negative_prompt,
                    height=height,
                    width=width
        )

        # return first generated PIL image
        return out.images[0]