Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,42 +1,50 @@
|
|
1 |
import gradio as gr
|
2 |
-
|
3 |
-
from all_models import models
|
4 |
-
from externalmod import gr_Interface_load, randomize_seed
|
5 |
import asyncio
|
6 |
import os
|
|
|
7 |
from threading import RLock
|
8 |
from pathlib import Path
|
|
|
|
|
9 |
|
10 |
# Create a lock for thread safety
|
11 |
lock = RLock()
|
12 |
|
13 |
-
# Load Hugging Face token from environment variable
|
14 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
15 |
|
16 |
-
# Function to load models
|
17 |
def load_fn(models):
|
18 |
global models_load
|
19 |
models_load = {}
|
|
|
20 |
for model in models:
|
21 |
if model not in models_load:
|
22 |
try:
|
23 |
print(f"Loading model: {model}")
|
24 |
-
m = gr_Interface_load(
|
|
|
|
|
|
|
|
|
|
|
25 |
models_load[model] = m
|
26 |
except Exception as e:
|
27 |
print(f"Error loading model {model}: {e}")
|
28 |
-
models_load[model] =
|
29 |
|
30 |
print("Loading models...")
|
31 |
load_fn(models)
|
32 |
print("Models loaded successfully.")
|
33 |
|
|
|
34 |
num_models = 1
|
35 |
starting_seed = randint(1941, 2024)
|
36 |
MAX_SEED = 3999999999
|
37 |
-
MAX_SEED = int(MAX_SEED)
|
38 |
inference_timeout = 600
|
39 |
|
|
|
40 |
def extend_choices(choices):
|
41 |
return choices[:num_models] + ['NA'] * (num_models - len(choices))
|
42 |
|
@@ -44,58 +52,69 @@ def update_imgbox(choices):
|
|
44 |
choices_extended = extend_choices(choices)
|
45 |
return [gr.Image(None, label=m, visible=(m != 'NA')) for m in choices_extended]
|
46 |
|
47 |
-
|
48 |
-
|
|
|
|
|
49 |
return None
|
50 |
-
|
51 |
kwargs = {"seed": seed}
|
52 |
try:
|
53 |
print(f"Running inference for model: {model_str} with prompt: '{prompt}'")
|
54 |
result = await asyncio.to_thread(models_load[model_str].fn, prompt=prompt, **kwargs, token=HF_TOKEN)
|
|
|
55 |
if result:
|
56 |
with lock:
|
57 |
png_path = "image.png"
|
58 |
result.save(png_path)
|
59 |
return str(Path(png_path).resolve())
|
|
|
|
|
60 |
except Exception as e:
|
61 |
print(f"Error during inference for {model_str}: {e}")
|
|
|
62 |
return None
|
63 |
|
|
|
64 |
def gen_fnseed(model_str, prompt, seed=1):
|
65 |
if model_str == 'NA':
|
66 |
return None
|
|
|
67 |
try:
|
68 |
loop = asyncio.new_event_loop()
|
69 |
asyncio.set_event_loop(loop)
|
70 |
-
result = loop.run_until_complete(infer(model_str, prompt, seed
|
71 |
except Exception as e:
|
72 |
print(f"Error generating image for {model_str}: {e}")
|
73 |
result = None
|
74 |
finally:
|
75 |
loop.close()
|
|
|
76 |
return result
|
77 |
|
|
|
78 |
print("Creating Gradio interface...")
|
79 |
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
80 |
gr.HTML("<center><h1>Compare-6</h1></center>")
|
|
|
81 |
with gr.Tab('Compare-6'):
|
82 |
txt_input = gr.Textbox(label='Your prompt:', lines=4)
|
83 |
gen_button = gr.Button('Generate up to 6 images')
|
84 |
seed = gr.Slider(label="Seed (0 to MAX)", minimum=0, maximum=MAX_SEED, value=starting_seed)
|
85 |
seed_rand = gr.Button("Randomize Seed 🎲")
|
86 |
-
|
87 |
seed_rand.click(randomize_seed, None, [seed], queue=False)
|
88 |
-
|
89 |
output = [gr.Image(label=m) for m in models[:num_models]]
|
90 |
current_models = [gr.Textbox(m, visible=False) for m in models[:num_models]]
|
91 |
-
|
92 |
for m, o in zip(current_models, output):
|
93 |
gen_button.click(gen_fnseed, inputs=[m, txt_input, seed], outputs=[o], queue=False)
|
94 |
-
|
95 |
with gr.Accordion('Model selection'):
|
96 |
model_choice = gr.CheckboxGroup(models, label=f'Choose up to {num_models} models')
|
97 |
model_choice.change(update_imgbox, model_choice, output)
|
98 |
model_choice.change(extend_choices, model_choice, current_models)
|
99 |
|
100 |
-
demo.queue(default_concurrency_limit=
|
101 |
demo.launch(show_api=False)
|
|
|
1 |
import gradio as gr
|
2 |
+
import torch
|
|
|
|
|
3 |
import asyncio
|
4 |
import os
|
5 |
+
from random import randint
|
6 |
from threading import RLock
|
7 |
from pathlib import Path
|
8 |
+
from all_models import models
|
9 |
+
from externalmod import gr_Interface_load, randomize_seed
|
10 |
|
11 |
# Create a lock for thread safety
|
12 |
lock = RLock()
|
13 |
|
14 |
+
# Load Hugging Face token from environment variable
|
15 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
16 |
|
17 |
+
# Function to load models with optimized settings
|
18 |
def load_fn(models):
|
19 |
global models_load
|
20 |
models_load = {}
|
21 |
+
|
22 |
for model in models:
|
23 |
if model not in models_load:
|
24 |
try:
|
25 |
print(f"Loading model: {model}")
|
26 |
+
m = gr_Interface_load(
|
27 |
+
f'models/{model}',
|
28 |
+
hf_token=HF_TOKEN,
|
29 |
+
torch_dtype=torch.float16 # Reduce memory usage
|
30 |
+
)
|
31 |
+
m.enable_model_cpu_offload() # Offload to CPU when not in use
|
32 |
models_load[model] = m
|
33 |
except Exception as e:
|
34 |
print(f"Error loading model {model}: {e}")
|
35 |
+
models_load[model] = None
|
36 |
|
37 |
print("Loading models...")
|
38 |
load_fn(models)
|
39 |
print("Models loaded successfully.")
|
40 |
|
41 |
+
# Constants
|
42 |
num_models = 1
|
43 |
starting_seed = randint(1941, 2024)
|
44 |
MAX_SEED = 3999999999
|
|
|
45 |
inference_timeout = 600
|
46 |
|
47 |
+
# Update UI components
|
48 |
def extend_choices(choices):
|
49 |
return choices[:num_models] + ['NA'] * (num_models - len(choices))
|
50 |
|
|
|
52 |
choices_extended = extend_choices(choices)
|
53 |
return [gr.Image(None, label=m, visible=(m != 'NA')) for m in choices_extended]
|
54 |
|
55 |
+
# Async inference function
|
56 |
+
async def infer(model_str, prompt, seed=1):
|
57 |
+
if model_str not in models_load or models_load[model_str] is None:
|
58 |
+
print(f"Model {model_str} is unavailable.")
|
59 |
return None
|
60 |
+
|
61 |
kwargs = {"seed": seed}
|
62 |
try:
|
63 |
print(f"Running inference for model: {model_str} with prompt: '{prompt}'")
|
64 |
result = await asyncio.to_thread(models_load[model_str].fn, prompt=prompt, **kwargs, token=HF_TOKEN)
|
65 |
+
|
66 |
if result:
|
67 |
with lock:
|
68 |
png_path = "image.png"
|
69 |
result.save(png_path)
|
70 |
return str(Path(png_path).resolve())
|
71 |
+
except torch.cuda.OutOfMemoryError:
|
72 |
+
print(f"CUDA memory error for {model_str}. Try reducing image size.")
|
73 |
except Exception as e:
|
74 |
print(f"Error during inference for {model_str}: {e}")
|
75 |
+
|
76 |
return None
|
77 |
|
78 |
+
# Synchronous wrapper
|
79 |
def gen_fnseed(model_str, prompt, seed=1):
|
80 |
if model_str == 'NA':
|
81 |
return None
|
82 |
+
|
83 |
try:
|
84 |
loop = asyncio.new_event_loop()
|
85 |
asyncio.set_event_loop(loop)
|
86 |
+
result = loop.run_until_complete(infer(model_str, prompt, seed))
|
87 |
except Exception as e:
|
88 |
print(f"Error generating image for {model_str}: {e}")
|
89 |
result = None
|
90 |
finally:
|
91 |
loop.close()
|
92 |
+
|
93 |
return result
|
94 |
|
95 |
+
# Gradio UI
|
96 |
print("Creating Gradio interface...")
|
97 |
with gr.Blocks(theme="Nymbo/Nymbo_Theme") as demo:
|
98 |
gr.HTML("<center><h1>Compare-6</h1></center>")
|
99 |
+
|
100 |
with gr.Tab('Compare-6'):
|
101 |
txt_input = gr.Textbox(label='Your prompt:', lines=4)
|
102 |
gen_button = gr.Button('Generate up to 6 images')
|
103 |
seed = gr.Slider(label="Seed (0 to MAX)", minimum=0, maximum=MAX_SEED, value=starting_seed)
|
104 |
seed_rand = gr.Button("Randomize Seed 🎲")
|
105 |
+
|
106 |
seed_rand.click(randomize_seed, None, [seed], queue=False)
|
107 |
+
|
108 |
output = [gr.Image(label=m) for m in models[:num_models]]
|
109 |
current_models = [gr.Textbox(m, visible=False) for m in models[:num_models]]
|
110 |
+
|
111 |
for m, o in zip(current_models, output):
|
112 |
gen_button.click(gen_fnseed, inputs=[m, txt_input, seed], outputs=[o], queue=False)
|
113 |
+
|
114 |
with gr.Accordion('Model selection'):
|
115 |
model_choice = gr.CheckboxGroup(models, label=f'Choose up to {num_models} models')
|
116 |
model_choice.change(update_imgbox, model_choice, output)
|
117 |
model_choice.change(extend_choices, model_choice, current_models)
|
118 |
|
119 |
+
demo.queue(default_concurrency_limit=20, max_size=50) # Adjusted for better stability
|
120 |
demo.launch(show_api=False)
|