Spaces:
Running
Running
Upload server.py with huggingface_hub
Browse files
server.py
ADDED
@@ -0,0 +1,72 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import sys
|
3 |
+
import base64
|
4 |
+
from io import BytesIO
|
5 |
+
sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
|
6 |
+
|
7 |
+
import torch
|
8 |
+
from torch import nn
|
9 |
+
from fastapi import FastAPI
|
10 |
+
import numpy as np
|
11 |
+
from PIL import Image
|
12 |
+
|
13 |
+
from dalle.models import Dalle
|
14 |
+
import logging
|
15 |
+
import streamlit as st
|
16 |
+
|
17 |
+
|
18 |
+
print("Loading models...")
|
19 |
+
app = FastAPI()
|
20 |
+
|
21 |
+
from huggingface_hub import hf_hub_download
|
22 |
+
|
23 |
+
logging.info("Start downloading")
|
24 |
+
full_dict_path = hf_hub_download(repo_id="ml6team/logo-generator", filename="full_dict_new.ckpt",
|
25 |
+
use_auth_token=st.secrets["model_download"])
|
26 |
+
logging.info("End downloading")
|
27 |
+
|
28 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
29 |
+
model = Dalle.from_pretrained("minDALL-E/1.3B")
|
30 |
+
|
31 |
+
model.load_state_dict(torch.load(full_dict_path, map_location=torch.device('cpu')))
|
32 |
+
model.to(device=device)
|
33 |
+
|
34 |
+
print("Models loaded !")
|
35 |
+
|
36 |
+
|
37 |
+
@app.get("/")
|
38 |
+
def read_root():
|
39 |
+
return {"minDALL-E!"}
|
40 |
+
|
41 |
+
|
42 |
+
@app.get("/{generate}")
|
43 |
+
def generate(prompt):
|
44 |
+
images = sample(prompt)
|
45 |
+
images = [to_base64(image) for image in images]
|
46 |
+
return {"images": images}
|
47 |
+
|
48 |
+
|
49 |
+
def sample(prompt):
|
50 |
+
# Sampling
|
51 |
+
logging.info("starting sampling")
|
52 |
+
images = (
|
53 |
+
model.sampling(prompt=prompt, top_k=96, top_p=None, softmax_temperature=1.0, num_candidates=9, device=device)
|
54 |
+
.cpu()
|
55 |
+
.numpy()
|
56 |
+
)
|
57 |
+
logging.info("sampling succeeded")
|
58 |
+
images = np.transpose(images, (0, 2, 3, 1))
|
59 |
+
|
60 |
+
|
61 |
+
pil_images = []
|
62 |
+
for i in range(len(images)):
|
63 |
+
im = Image.fromarray((images[i] * 255).astype(np.uint8))
|
64 |
+
pil_images.append(im)
|
65 |
+
|
66 |
+
return pil_images
|
67 |
+
|
68 |
+
|
69 |
+
def to_base64(pil_image):
|
70 |
+
buffered = BytesIO()
|
71 |
+
pil_image.save(buffered, format="JPEG")
|
72 |
+
return base64.b64encode(buffered.getvalue())
|