Update app.py
Browse files
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import numpy as np
|
2 |
from PIL import Image
|
3 |
-
from huggingface_hub import snapshot_download
|
4 |
from leffa.transform import LeffaTransform
|
5 |
from leffa.model import LeffaModel
|
6 |
from leffa.inference import LeffaInference
|
@@ -13,11 +13,19 @@ from diffusers import DiffusionPipeline
|
|
13 |
from transformers import pipeline
|
14 |
import gradio as gr
|
15 |
import os
|
16 |
-
from huggingface_hub import login
|
17 |
import random
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
|
19 |
# 상수 정의
|
20 |
MAX_SEED = 2**32 - 1
|
|
|
|
|
|
|
21 |
|
22 |
# Hugging Face 토큰 설정 및 로그인
|
23 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
@@ -25,20 +33,26 @@ if HF_TOKEN is None:
|
|
25 |
raise ValueError("Please set the HF_TOKEN environment variable")
|
26 |
login(token=HF_TOKEN)
|
27 |
|
28 |
-
#
|
29 |
-
|
30 |
-
|
31 |
-
|
|
|
|
|
|
|
32 |
|
33 |
-
#
|
|
|
|
|
|
|
34 |
fashion_pipe = DiffusionPipeline.from_pretrained(
|
35 |
-
|
36 |
-
torch_dtype=torch.
|
37 |
use_auth_token=HF_TOKEN
|
38 |
)
|
39 |
-
fashion_pipe.
|
40 |
|
41 |
-
# 번역기 초기화
|
42 |
translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")
|
43 |
|
44 |
# Leffa 체크포인트 다운로드
|
@@ -55,18 +69,22 @@ densepose_predictor = DensePosePredictor(
|
|
55 |
weights_path="./ckpts/densepose/model_final_162be9.pkl",
|
56 |
)
|
57 |
|
|
|
58 |
vt_model = LeffaModel(
|
59 |
pretrained_model_name_or_path="./ckpts/stable-diffusion-inpainting",
|
60 |
pretrained_model="./ckpts/virtual_tryon.pth",
|
|
|
61 |
)
|
62 |
vt_inference = LeffaInference(model=vt_model)
|
63 |
|
64 |
pt_model = LeffaModel(
|
65 |
pretrained_model_name_or_path="./ckpts/stable-diffusion-xl-1.0-inpainting-0.1",
|
66 |
pretrained_model="./ckpts/pose_transfer.pth",
|
|
|
67 |
)
|
68 |
pt_inference = LeffaInference(model=pt_model)
|
69 |
|
|
|
70 |
def contains_korean(text):
|
71 |
return any(ord('가') <= ord(char) <= ord('힣') for char in text)
|
72 |
|
|
|
1 |
import numpy as np
|
2 |
from PIL import Image
|
3 |
+
from huggingface_hub import snapshot_download, login
|
4 |
from leffa.transform import LeffaTransform
|
5 |
from leffa.model import LeffaModel
|
6 |
from leffa.inference import LeffaInference
|
|
|
13 |
from transformers import pipeline
|
14 |
import gradio as gr
|
15 |
import os
|
|
|
16 |
import random
|
17 |
+
import gc
|
18 |
+
|
19 |
+
# 메모리 최적화 설정
|
20 |
+
torch.backends.cudnn.benchmark = True
|
21 |
+
torch.backends.cuda.matmul.allow_tf32 = True
|
22 |
+
os.environ['PYTORCH_CUDA_ALLOC_CONF'] = 'max_split_size_mb:512'
|
23 |
|
24 |
# 상수 정의
|
25 |
MAX_SEED = 2**32 - 1
|
26 |
+
BASE_MODEL = "black-forest-labs/FLUX.1-dev"
|
27 |
+
MODEL_LORA_REPO = "Motas/Flux_Fashion_Photography_Style"
|
28 |
+
CLOTHES_LORA_REPO = "prithivMLmods/Canopus-Clothing-Flux-LoRA"
|
29 |
|
30 |
# Hugging Face 토큰 설정 및 로그인
|
31 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
|
|
33 |
raise ValueError("Please set the HF_TOKEN environment variable")
|
34 |
login(token=HF_TOKEN)
|
35 |
|
36 |
+
# 메모리 정리 함수
|
37 |
+
def clear_memory():
|
38 |
+
torch.cuda.empty_cache()
|
39 |
+
gc.collect()
|
40 |
+
|
41 |
+
# 초기 메모리 정리
|
42 |
+
clear_memory()
|
43 |
|
44 |
+
# CUDA 사용 가능 여부 확인
|
45 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
46 |
+
|
47 |
+
# FLUX 모델 초기화
|
48 |
fashion_pipe = DiffusionPipeline.from_pretrained(
|
49 |
+
BASE_MODEL,
|
50 |
+
torch_dtype=torch.float16,
|
51 |
use_auth_token=HF_TOKEN
|
52 |
)
|
53 |
+
fashion_pipe.enable_model_cpu_offload()
|
54 |
|
55 |
+
# 번역기 초기화
|
56 |
translator = pipeline("translation", model="Helsinki-NLP/opus-mt-ko-en")
|
57 |
|
58 |
# Leffa 체크포인트 다운로드
|
|
|
69 |
weights_path="./ckpts/densepose/model_final_162be9.pkl",
|
70 |
)
|
71 |
|
72 |
+
# Leffa 모델 초기화
|
73 |
vt_model = LeffaModel(
|
74 |
pretrained_model_name_or_path="./ckpts/stable-diffusion-inpainting",
|
75 |
pretrained_model="./ckpts/virtual_tryon.pth",
|
76 |
+
device_map="auto"
|
77 |
)
|
78 |
vt_inference = LeffaInference(model=vt_model)
|
79 |
|
80 |
pt_model = LeffaModel(
|
81 |
pretrained_model_name_or_path="./ckpts/stable-diffusion-xl-1.0-inpainting-0.1",
|
82 |
pretrained_model="./ckpts/pose_transfer.pth",
|
83 |
+
device_map="auto"
|
84 |
)
|
85 |
pt_inference = LeffaInference(model=pt_model)
|
86 |
|
87 |
+
|
88 |
def contains_korean(text):
|
89 |
return any(ord('가') <= ord(char) <= ord('힣') for char in text)
|
90 |
|