Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -12,6 +12,11 @@ from gradio_imageslider import ImageSlider
|
|
12 |
from PIL import Image
|
13 |
from huggingface_hub import snapshot_download
|
14 |
|
|
|
|
|
|
|
|
|
|
|
15 |
css = """
|
16 |
#col-container {
|
17 |
margin: 0 auto;
|
@@ -19,97 +24,96 @@ css = """
|
|
19 |
}
|
20 |
"""
|
21 |
|
22 |
-
# Device
|
23 |
if torch.cuda.is_available():
|
24 |
power_device = "GPU"
|
25 |
device = "cuda"
|
26 |
-
dtype = torch.float16 #
|
|
|
|
|
27 |
else:
|
28 |
power_device = "CPU"
|
29 |
device = "cpu"
|
30 |
dtype = torch.float32
|
31 |
|
32 |
-
# Reduce CUDA memory usage
|
33 |
-
torch.cuda.empty_cache()
|
34 |
-
if torch.cuda.is_available():
|
35 |
-
torch.cuda.set_per_process_memory_fraction(0.7) # Use only 70% of GPU memory
|
36 |
-
|
37 |
huggingface_token = os.getenv("HUGGINFACE_TOKEN")
|
38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
model_path = snapshot_download(
|
40 |
repo_id="black-forest-labs/FLUX.1-dev",
|
41 |
repo_type="model",
|
42 |
-
ignore_patterns=["*.md", "*..gitattributes"],
|
43 |
local_dir="FLUX.1-dev",
|
44 |
token=huggingface_token,
|
45 |
)
|
46 |
|
47 |
-
# Load
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
low_cpu_mem_usage=True,
|
60 |
-
use_safetensors=True
|
61 |
-
)
|
62 |
|
63 |
-
# Enable all
|
64 |
-
pipe.enable_model_cpu_offload()
|
65 |
-
pipe.enable_attention_slicing(1)
|
66 |
-
pipe.enable_sequential_cpu_offload()
|
67 |
-
pipe.enable_vae_slicing()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
-
#
|
70 |
MAX_SEED = 1000000
|
71 |
-
MAX_PIXEL_BUDGET =
|
72 |
|
73 |
def check_resources():
|
74 |
if torch.cuda.is_available():
|
75 |
-
gpu_memory = torch.cuda.get_device_properties(0).total_memory
|
76 |
memory_allocated = torch.cuda.memory_allocated(0)
|
77 |
-
|
78 |
-
|
|
|
|
|
79 |
return True
|
80 |
|
81 |
def process_input(input_image, upscale_factor, **kwargs):
|
82 |
-
# Convert image to RGB mode to ensure compatibility
|
83 |
input_image = input_image.convert('RGB')
|
84 |
|
|
|
85 |
w, h = input_image.size
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
)
|
95 |
-
|
96 |
-
f"Resizing input image to fit memory constraints..."
|
97 |
-
)
|
98 |
-
input_image = input_image.resize(
|
99 |
-
(
|
100 |
-
int(aspect_ratio * MAX_PIXEL_BUDGET**0.5 // upscale_factor),
|
101 |
-
int(MAX_PIXEL_BUDGET**0.5 // aspect_ratio // upscale_factor),
|
102 |
-
),
|
103 |
-
Image.LANCZOS
|
104 |
-
)
|
105 |
-
was_resized = True
|
106 |
-
|
107 |
-
# resize to multiple of 8
|
108 |
w, h = input_image.size
|
109 |
w = w - w % 8
|
110 |
h = h - h % 8
|
111 |
-
|
112 |
-
return input_image.resize((w, h)),
|
113 |
|
114 |
@spaces.GPU
|
115 |
def infer(
|
@@ -122,55 +126,32 @@ def infer(
|
|
122 |
progress=gr.Progress(track_tqdm=True),
|
123 |
):
|
124 |
try:
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
# Clear CUDA cache before processing
|
130 |
-
if device == "cuda":
|
131 |
-
torch.cuda.empty_cache()
|
132 |
-
|
133 |
if randomize_seed:
|
134 |
seed = random.randint(0, MAX_SEED)
|
135 |
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
# rescale with upscale factor
|
142 |
-
w, h = input_image.size
|
143 |
-
control_image = input_image.resize((w * upscale_factor, h * upscale_factor))
|
144 |
-
|
145 |
-
generator = torch.Generator().manual_seed(seed)
|
146 |
-
|
147 |
-
gr.Info("Upscaling image...")
|
148 |
-
with torch.inference_mode(): # Use inference mode to save memory
|
149 |
image = pipe(
|
150 |
prompt="",
|
151 |
-
control_image=
|
152 |
controlnet_conditioning_scale=controlnet_conditioning_scale,
|
153 |
num_inference_steps=num_inference_steps,
|
154 |
-
guidance_scale=3.5
|
155 |
-
height=
|
156 |
-
width=
|
157 |
generator=generator,
|
158 |
).images[0]
|
|
|
|
|
|
|
|
|
|
|
159 |
|
160 |
-
if was_resized:
|
161 |
-
gr.Info(
|
162 |
-
f"Resizing output image to final size..."
|
163 |
-
)
|
164 |
-
|
165 |
-
# resize to target desired size
|
166 |
-
image = image.resize((w_original * upscale_factor, h_original * upscale_factor))
|
167 |
-
return [true_input_image, image, seed]
|
168 |
-
|
169 |
-
except RuntimeError as e:
|
170 |
-
if "out of memory" in str(e):
|
171 |
-
gr.Warning("Not enough GPU memory. Try reducing the upscale factor or image size.")
|
172 |
-
return None
|
173 |
-
raise e
|
174 |
except Exception as e:
|
175 |
gr.Error(f"An error occurred: {str(e)}")
|
176 |
return None
|
@@ -184,25 +165,25 @@ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
|
184 |
input_im = gr.Image(label="Input Image", type="pil")
|
185 |
with gr.Column(scale=1):
|
186 |
num_inference_steps = gr.Slider(
|
187 |
-
label="
|
188 |
-
minimum=
|
189 |
-
maximum=
|
190 |
step=1,
|
191 |
-
value=
|
192 |
)
|
193 |
upscale_factor = gr.Slider(
|
194 |
-
label="
|
195 |
minimum=1,
|
196 |
-
maximum=
|
197 |
step=1,
|
198 |
-
value=1,
|
199 |
)
|
200 |
controlnet_conditioning_scale = gr.Slider(
|
201 |
-
label="
|
202 |
minimum=0.1,
|
203 |
-
maximum=
|
204 |
step=0.1,
|
205 |
-
value=0.
|
206 |
)
|
207 |
seed = gr.Slider(
|
208 |
label="Seed",
|
@@ -211,18 +192,17 @@ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
|
211 |
step=1,
|
212 |
value=42,
|
213 |
)
|
214 |
-
|
215 |
-
randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
|
216 |
|
217 |
with gr.Row():
|
218 |
-
result = ImageSlider(label="
|
219 |
|
220 |
current_dir = os.path.dirname(os.path.abspath(__file__))
|
221 |
|
222 |
examples = gr.Examples(
|
223 |
examples=[
|
224 |
-
[42, False, os.path.join(current_dir, "z1.webp"),
|
225 |
-
[42, False, os.path.join(current_dir, "z2.webp"),
|
226 |
],
|
227 |
inputs=[
|
228 |
seed,
|
@@ -234,7 +214,7 @@ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
|
234 |
],
|
235 |
fn=infer,
|
236 |
outputs=result,
|
237 |
-
cache_examples=
|
238 |
)
|
239 |
|
240 |
gr.on(
|
@@ -252,11 +232,13 @@ with gr.Blocks(theme="Yntec/HaleyCH_Theme_Orange", css=css) as demo:
|
|
252 |
show_api=False,
|
253 |
)
|
254 |
|
255 |
-
# Launch with minimal
|
256 |
demo.queue(max_size=1).launch(
|
257 |
share=False,
|
258 |
debug=True,
|
259 |
show_error=True,
|
260 |
max_threads=1,
|
261 |
-
enable_queue=True
|
|
|
|
|
262 |
)
|
|
|
12 |
from PIL import Image
|
13 |
from huggingface_hub import snapshot_download
|
14 |
|
15 |
+
# 메모리 관리를 위한 gc 추가
|
16 |
+
import gc
|
17 |
+
gc.collect()
|
18 |
+
torch.cuda.empty_cache()
|
19 |
+
|
20 |
css = """
|
21 |
#col-container {
|
22 |
margin: 0 auto;
|
|
|
24 |
}
|
25 |
"""
|
26 |
|
27 |
+
# Device setup with minimal memory usage
|
28 |
if torch.cuda.is_available():
|
29 |
power_device = "GPU"
|
30 |
device = "cuda"
|
31 |
+
dtype = torch.float16 # Use float16 for minimum memory
|
32 |
+
# Set CUDA memory fraction to 50%
|
33 |
+
torch.cuda.set_per_process_memory_fraction(0.5)
|
34 |
else:
|
35 |
power_device = "CPU"
|
36 |
device = "cpu"
|
37 |
dtype = torch.float32
|
38 |
|
|
|
|
|
|
|
|
|
|
|
39 |
huggingface_token = os.getenv("HUGGINFACE_TOKEN")
|
40 |
|
41 |
+
# Minimal model configuration
|
42 |
+
model_config = {
|
43 |
+
"low_cpu_mem_usage": True,
|
44 |
+
"torch_dtype": dtype,
|
45 |
+
"use_safetensors": True,
|
46 |
+
"variant": "fp16", # Use fp16 variant if available
|
47 |
+
}
|
48 |
+
|
49 |
model_path = snapshot_download(
|
50 |
repo_id="black-forest-labs/FLUX.1-dev",
|
51 |
repo_type="model",
|
52 |
+
ignore_patterns=["*.md", "*..gitattributes", "*.bin"], # Ignore unnecessary files
|
53 |
local_dir="FLUX.1-dev",
|
54 |
token=huggingface_token,
|
55 |
)
|
56 |
|
57 |
+
# Load models with minimal configuration
|
58 |
+
try:
|
59 |
+
controlnet = FluxControlNetModel.from_pretrained(
|
60 |
+
"jasperai/Flux.1-dev-Controlnet-Upscaler",
|
61 |
+
**model_config
|
62 |
+
).to(device)
|
63 |
+
|
64 |
+
pipe = FluxControlNetPipeline.from_pretrained(
|
65 |
+
model_path,
|
66 |
+
controlnet=controlnet,
|
67 |
+
**model_config
|
68 |
+
)
|
|
|
|
|
|
|
69 |
|
70 |
+
# Enable all memory optimizations
|
71 |
+
pipe.enable_model_cpu_offload()
|
72 |
+
pipe.enable_attention_slicing(1)
|
73 |
+
pipe.enable_sequential_cpu_offload()
|
74 |
+
pipe.enable_vae_slicing()
|
75 |
+
|
76 |
+
# Clear memory after loading
|
77 |
+
gc.collect()
|
78 |
+
torch.cuda.empty_cache()
|
79 |
+
|
80 |
+
except Exception as e:
|
81 |
+
print(f"Error loading models: {e}")
|
82 |
+
raise
|
83 |
|
84 |
+
# Extremely reduced parameters
|
85 |
MAX_SEED = 1000000
|
86 |
+
MAX_PIXEL_BUDGET = 128 * 128 # Extremely reduced from 256 * 256
|
87 |
|
88 |
def check_resources():
|
89 |
if torch.cuda.is_available():
|
|
|
90 |
memory_allocated = torch.cuda.memory_allocated(0)
|
91 |
+
memory_reserved = torch.cuda.memory_reserved(0)
|
92 |
+
if memory_allocated/memory_reserved > 0.7: # 70% threshold
|
93 |
+
gc.collect()
|
94 |
+
torch.cuda.empty_cache()
|
95 |
return True
|
96 |
|
97 |
def process_input(input_image, upscale_factor, **kwargs):
|
|
|
98 |
input_image = input_image.convert('RGB')
|
99 |
|
100 |
+
# Reduce image size more aggressively
|
101 |
w, h = input_image.size
|
102 |
+
max_size = int(np.sqrt(MAX_PIXEL_BUDGET))
|
103 |
+
if w > max_size or h > max_size:
|
104 |
+
if w > h:
|
105 |
+
new_w = max_size
|
106 |
+
new_h = int(h * max_size / w)
|
107 |
+
else:
|
108 |
+
new_h = max_size
|
109 |
+
new_w = int(w * max_size / h)
|
110 |
+
input_image = input_image.resize((new_w, new_h), Image.LANCZOS)
|
111 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
112 |
w, h = input_image.size
|
113 |
w = w - w % 8
|
114 |
h = h - h % 8
|
115 |
+
|
116 |
+
return input_image.resize((w, h)), w, h, True
|
117 |
|
118 |
@spaces.GPU
|
119 |
def infer(
|
|
|
126 |
progress=gr.Progress(track_tqdm=True),
|
127 |
):
|
128 |
try:
|
129 |
+
gc.collect()
|
130 |
+
torch.cuda.empty_cache()
|
131 |
+
|
|
|
|
|
|
|
|
|
|
|
132 |
if randomize_seed:
|
133 |
seed = random.randint(0, MAX_SEED)
|
134 |
|
135 |
+
input_image, w, h, _ = process_input(input_image, upscale_factor)
|
136 |
+
|
137 |
+
with torch.inference_mode():
|
138 |
+
generator = torch.Generator().manual_seed(seed)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
139 |
image = pipe(
|
140 |
prompt="",
|
141 |
+
control_image=input_image,
|
142 |
controlnet_conditioning_scale=controlnet_conditioning_scale,
|
143 |
num_inference_steps=num_inference_steps,
|
144 |
+
guidance_scale=2.0, # Reduced from 3.5
|
145 |
+
height=h,
|
146 |
+
width=w,
|
147 |
generator=generator,
|
148 |
).images[0]
|
149 |
+
|
150 |
+
gc.collect()
|
151 |
+
torch.cuda.empty_cache()
|
152 |
+
|
153 |
+
return [input_image, image, seed]
|
154 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
155 |
except Exception as e:
|
156 |
gr.Error(f"An error occurred: {str(e)}")
|
157 |
return None
|
|
|
165 |
input_im = gr.Image(label="Input Image", type="pil")
|
166 |
with gr.Column(scale=1):
|
167 |
num_inference_steps = gr.Slider(
|
168 |
+
label="Steps",
|
169 |
+
minimum=1,
|
170 |
+
maximum=20, # Reduced from 30
|
171 |
step=1,
|
172 |
+
value=10, # Reduced from 20
|
173 |
)
|
174 |
upscale_factor = gr.Slider(
|
175 |
+
label="Scale",
|
176 |
minimum=1,
|
177 |
+
maximum=1, # Fixed at 1
|
178 |
step=1,
|
179 |
+
value=1,
|
180 |
)
|
181 |
controlnet_conditioning_scale = gr.Slider(
|
182 |
+
label="Control Scale",
|
183 |
minimum=0.1,
|
184 |
+
maximum=0.5, # Reduced from 1.0
|
185 |
step=0.1,
|
186 |
+
value=0.3, # Reduced from 0.5
|
187 |
)
|
188 |
seed = gr.Slider(
|
189 |
label="Seed",
|
|
|
192 |
step=1,
|
193 |
value=42,
|
194 |
)
|
195 |
+
randomize_seed = gr.Checkbox(label="Random Seed", value=True)
|
|
|
196 |
|
197 |
with gr.Row():
|
198 |
+
result = ImageSlider(label="Result", type="pil", interactive=True)
|
199 |
|
200 |
current_dir = os.path.dirname(os.path.abspath(__file__))
|
201 |
|
202 |
examples = gr.Examples(
|
203 |
examples=[
|
204 |
+
[42, False, os.path.join(current_dir, "z1.webp"), 10, 1, 0.3],
|
205 |
+
[42, False, os.path.join(current_dir, "z2.webp"), 10, 1, 0.3],
|
206 |
],
|
207 |
inputs=[
|
208 |
seed,
|
|
|
214 |
],
|
215 |
fn=infer,
|
216 |
outputs=result,
|
217 |
+
cache_examples=False, # Disable caching
|
218 |
)
|
219 |
|
220 |
gr.on(
|
|
|
232 |
show_api=False,
|
233 |
)
|
234 |
|
235 |
+
# Launch with minimal resources
|
236 |
demo.queue(max_size=1).launch(
|
237 |
share=False,
|
238 |
debug=True,
|
239 |
show_error=True,
|
240 |
max_threads=1,
|
241 |
+
enable_queue=True,
|
242 |
+
cache_examples=False,
|
243 |
+
quiet=True,
|
244 |
)
|