Spaces:
Running
on
A100
Running
on
A100
Commit
·
b24c174
1
Parent(s):
def3000
Add multiprocessing
Browse files
app.py
CHANGED
@@ -8,6 +8,7 @@ from dataclasses import fields
|
|
8 |
from urllib.request import urlretrieve
|
9 |
|
10 |
import gradio as gr
|
|
|
11 |
import transformers
|
12 |
from legogpt.models import LegoGPT, LegoGPTConfig
|
13 |
|
@@ -39,47 +40,8 @@ def main():
|
|
39 |
print('Running in Hugging Face Space, setting up environment...')
|
40 |
setup()
|
41 |
|
42 |
-
model_cfg = LegoGPTConfig(max_regenerations=
|
43 |
-
|
44 |
-
|
45 |
-
def generate_lego(
|
46 |
-
prompt: str,
|
47 |
-
temperature: float | None,
|
48 |
-
seed: int | None,
|
49 |
-
max_bricks: int | None,
|
50 |
-
max_brick_rejections: int | None,
|
51 |
-
max_regenerations: int | None,
|
52 |
-
):
|
53 |
-
# Set model parameters
|
54 |
-
if temperature is not None: model.temperature = temperature
|
55 |
-
if max_bricks is not None: model.max_bricks = max_bricks
|
56 |
-
if max_brick_rejections is not None: model.max_brick_rejections = max_brick_rejections
|
57 |
-
if max_regenerations is not None: model.max_regenerations = max_regenerations
|
58 |
-
if seed is not None: transformers.set_seed(seed)
|
59 |
-
|
60 |
-
# Generate LEGO
|
61 |
-
print(f'Generating LEGO for prompt: "{prompt}"')
|
62 |
-
start_time = time.time()
|
63 |
-
output = model(prompt)
|
64 |
-
|
65 |
-
# Write output LDR to file
|
66 |
-
output_dir = os.path.abspath('out')
|
67 |
-
output_uuid = str(uuid.uuid4())
|
68 |
-
os.makedirs(output_dir, exist_ok=True)
|
69 |
-
ldr_filename = os.path.join(output_dir, f'{output_uuid}.ldr')
|
70 |
-
with open(ldr_filename, 'w') as f:
|
71 |
-
f.write(output['lego'].to_ldr())
|
72 |
-
print(f'Finished generation in {time.time() - start_time:.1f}s!')
|
73 |
-
|
74 |
-
# Render LEGO model to image
|
75 |
-
print('Rendering image...')
|
76 |
-
start_time = time.time()
|
77 |
-
img_filename = os.path.join(output_dir, f'{output_uuid}.png')
|
78 |
-
subprocess.run(['python', 'render_lego.py', '--in_file', ldr_filename, '--out_file', img_filename],
|
79 |
-
check=True) # Run render as a subprocess to prevent issues with Blender
|
80 |
-
print(f'Finished rendering in {time.time() - start_time:.1f}s!')
|
81 |
-
|
82 |
-
return img_filename, output['lego']
|
83 |
|
84 |
# Define inputs and outputs
|
85 |
in_prompt = gr.Textbox(label='Prompt', placeholder='Enter a prompt to generate a LEGO model.')
|
@@ -99,7 +61,7 @@ def main():
|
|
99 |
|
100 |
# Define Gradio interface
|
101 |
demo = gr.Interface(
|
102 |
-
fn=
|
103 |
title='LegoGPT Demo',
|
104 |
description='Official demo for [LegoGPT](https://avalovelace1.github.io/LegoGPT/), the first approach for generating physically stable LEGO brick models from text prompts.\n\n'
|
105 |
'The model is restricted to creating structures made of 1-unit-tall cuboid bricks on a 20x20x20 grid. It was trained on a dataset of 21 object categories: '
|
@@ -123,9 +85,65 @@ def main():
|
|
123 |
fn=lambda *args: (args[-1], examples[args[0]]['output_txt']),
|
124 |
run_on_click=True,
|
125 |
)
|
|
|
|
|
|
|
126 |
demo.launch(share=True)
|
127 |
|
128 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
129 |
def get_help_string(field_name: str) -> str:
|
130 |
"""
|
131 |
:param field_name: Name of a field in LegoGPTConfig.
|
|
|
8 |
from urllib.request import urlretrieve
|
9 |
|
10 |
import gradio as gr
|
11 |
+
import torch.multiprocessing as mp
|
12 |
import transformers
|
13 |
from legogpt.models import LegoGPT, LegoGPTConfig
|
14 |
|
|
|
40 |
print('Running in Hugging Face Space, setting up environment...')
|
41 |
setup()
|
42 |
|
43 |
+
model_cfg = LegoGPTConfig(max_regenerations=5)
|
44 |
+
generator = LegoGenerator(LegoGPT(model_cfg))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
# Define inputs and outputs
|
47 |
in_prompt = gr.Textbox(label='Prompt', placeholder='Enter a prompt to generate a LEGO model.')
|
|
|
61 |
|
62 |
# Define Gradio interface
|
63 |
demo = gr.Interface(
|
64 |
+
fn=generator.generate_lego_subprocess,
|
65 |
title='LegoGPT Demo',
|
66 |
description='Official demo for [LegoGPT](https://avalovelace1.github.io/LegoGPT/), the first approach for generating physically stable LEGO brick models from text prompts.\n\n'
|
67 |
'The model is restricted to creating structures made of 1-unit-tall cuboid bricks on a 20x20x20 grid. It was trained on a dataset of 21 object categories: '
|
|
|
85 |
fn=lambda *args: (args[-1], examples[args[0]]['output_txt']),
|
86 |
run_on_click=True,
|
87 |
)
|
88 |
+
|
89 |
+
concurrency_limit = 2 if os.environ.get('CONCURRENCY_LIMIT') is None else int(os.environ.get('CONCURRENCY_LIMIT'))
|
90 |
+
demo.queue(default_concurrency_limit=concurrency_limit)
|
91 |
demo.launch(share=True)
|
92 |
|
93 |
|
94 |
+
class LegoGenerator:
|
95 |
+
def __init__(self, model: LegoGPT):
|
96 |
+
self.model = model
|
97 |
+
self.ctx = mp.get_context('spawn')
|
98 |
+
|
99 |
+
def generate_lego(
|
100 |
+
self,
|
101 |
+
prompt: str,
|
102 |
+
temperature: float | None,
|
103 |
+
seed: int | None,
|
104 |
+
max_bricks: int | None,
|
105 |
+
max_brick_rejections: int | None,
|
106 |
+
max_regenerations: int | None,
|
107 |
+
):
|
108 |
+
# Set model parameters
|
109 |
+
if temperature is not None: self.model.temperature = temperature
|
110 |
+
if max_bricks is not None: self.model.max_bricks = max_bricks
|
111 |
+
if max_brick_rejections is not None: self.model.max_brick_rejections = max_brick_rejections
|
112 |
+
if max_regenerations is not None: self.model.max_regenerations = max_regenerations
|
113 |
+
if seed is not None: transformers.set_seed(seed)
|
114 |
+
|
115 |
+
# Generate LEGO
|
116 |
+
print(f'Generating LEGO for prompt: "{prompt}"')
|
117 |
+
start_time = time.time()
|
118 |
+
output = self.model(prompt)
|
119 |
+
|
120 |
+
# Write output LDR to file
|
121 |
+
output_dir = os.path.abspath('out')
|
122 |
+
output_uuid = str(uuid.uuid4())
|
123 |
+
os.makedirs(output_dir, exist_ok=True)
|
124 |
+
ldr_filename = os.path.join(output_dir, f'{output_uuid}.ldr')
|
125 |
+
with open(ldr_filename, 'w') as f:
|
126 |
+
f.write(output['lego'].to_ldr())
|
127 |
+
print(f'Finished generation in {time.time() - start_time:.1f}s!')
|
128 |
+
|
129 |
+
# Render LEGO model to image
|
130 |
+
print('Rendering image...')
|
131 |
+
start_time = time.time()
|
132 |
+
img_filename = os.path.join(output_dir, f'{output_uuid}.png')
|
133 |
+
subprocess.run(['python', 'render_lego.py', '--in_file', ldr_filename, '--out_file', img_filename],
|
134 |
+
check=True) # Run render as a subprocess to prevent issues with Blender
|
135 |
+
print(f'Finished rendering in {time.time() - start_time:.1f}s!')
|
136 |
+
|
137 |
+
return img_filename, output['lego']
|
138 |
+
|
139 |
+
def generate_lego_subprocess(self, *args):
|
140 |
+
"""
|
141 |
+
Run generation as a subprocess so that multiple requests can be handled concurrently.
|
142 |
+
"""
|
143 |
+
with self.ctx.Pool(1) as pool:
|
144 |
+
return pool.starmap(self.generate_lego, [args])[0]
|
145 |
+
|
146 |
+
|
147 |
def get_help_string(field_name: str) -> str:
|
148 |
"""
|
149 |
:param field_name: Name of a field in LegoGPTConfig.
|