zhiweili
commited on
Commit
Β·
fc1393b
1
Parent(s):
688a239
add app_tensorrt
Browse files- app.py +1 -1
- app_tensorrt.py +85 -0
- requirements.txt +4 -9
app.py
CHANGED
@@ -1,6 +1,6 @@
|
|
1 |
import gradio as gr
|
2 |
|
3 |
-
from
|
4 |
|
5 |
with gr.Blocks(css="style.css") as demo:
|
6 |
with gr.Tabs():
|
|
|
1 |
import gradio as gr
|
2 |
|
3 |
+
from app_tensorrt import create_demo as create_demo_face
|
4 |
|
5 |
with gr.Blocks(css="style.css") as demo:
|
6 |
with gr.Tabs():
|
app_tensorrt.py
ADDED
@@ -0,0 +1,85 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch_tensorrt
|
3 |
+
|
4 |
+
from diffusers import (
|
5 |
+
DDPMScheduler,
|
6 |
+
StableDiffusionXLImg2ImgPipeline,
|
7 |
+
AutoencoderKL,
|
8 |
+
)
|
9 |
+
|
10 |
+
BASE_MODEL = "stabilityai/stable-diffusion-xl-base-1.0"
|
11 |
+
device = "cuda"
|
12 |
+
|
13 |
+
vae = AutoencoderKL.from_pretrained(
|
14 |
+
"madebyollin/sdxl-vae-fp16-fix",
|
15 |
+
torch_dtype=torch.float16,
|
16 |
+
)
|
17 |
+
|
18 |
+
base_pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
|
19 |
+
BASE_MODEL,
|
20 |
+
vae=vae,
|
21 |
+
torch_dtype=torch.float16,
|
22 |
+
variant="fp16",
|
23 |
+
use_safetensors=True,
|
24 |
+
)
|
25 |
+
base_pipe = base_pipe.to(device, silence_dtype_warnings=True)
|
26 |
+
base_pipe.scheduler = DDPMScheduler.from_pretrained(
|
27 |
+
BASE_MODEL,
|
28 |
+
subfolder="scheduler",
|
29 |
+
)
|
30 |
+
|
31 |
+
backend = "torch_tensorrt"
|
32 |
+
|
33 |
+
|
34 |
+
|
35 |
+
# print('Loading compiled model...')
|
36 |
+
# loadedModel = torch_tensorrt.load("compiled_pipe.ep").module()
|
37 |
+
# print('Compiled model loaded!')
|
38 |
+
|
39 |
+
|
40 |
+
|
41 |
+
def create_demo() -> gr.Blocks:
|
42 |
+
|
43 |
+
@spaces.GPU(duration=30)
|
44 |
+
def text_to_image(
|
45 |
+
prompt:str,
|
46 |
+
steps:int,
|
47 |
+
):
|
48 |
+
print('Compiling model...')
|
49 |
+
compiledModel = torch.compile(
|
50 |
+
base_pipe.unet,
|
51 |
+
backend=backend,
|
52 |
+
options={
|
53 |
+
"truncate_long_and_double": True,
|
54 |
+
"enabled_precisions": {torch.float32, torch.float16},
|
55 |
+
},
|
56 |
+
dynamic=False,
|
57 |
+
)
|
58 |
+
print('Model compiled!')
|
59 |
+
|
60 |
+
print('Saving compiled model...')
|
61 |
+
torch_tensorrt.save(compiledModel, "compiled_pipe.ep")
|
62 |
+
print('Compiled model saved!')
|
63 |
+
|
64 |
+
with gr.Blocks() as demo:
|
65 |
+
with gr.Row():
|
66 |
+
with gr.Column():
|
67 |
+
prompt = gr.Textbox(label="Prompt", placeholder="Write a prompt here", lines=2, value="A beautiful sunset over the city")
|
68 |
+
with gr.Column():
|
69 |
+
steps = gr.Slider(minimum=1, maximum=100, value=5, step=1, label="Num Steps")
|
70 |
+
g_btn = gr.Button("Generate")
|
71 |
+
|
72 |
+
with gr.Row():
|
73 |
+
with gr.Column():
|
74 |
+
generated_image = gr.Image(label="Generated Image", type="pil", interactive=False)
|
75 |
+
with gr.Column():
|
76 |
+
time_cost = gr.Textbox(label="Time Cost", lines=1, interactive=False)
|
77 |
+
|
78 |
+
g_btn.click(
|
79 |
+
fn=text_to_image,
|
80 |
+
inputs=[prompt, steps],
|
81 |
+
# outputs=[generated_image, time_cost],
|
82 |
+
outputs=[],
|
83 |
+
)
|
84 |
+
|
85 |
+
return demo
|
requirements.txt
CHANGED
@@ -1,13 +1,8 @@
|
|
1 |
gradio
|
2 |
-
torch
|
3 |
-
|
|
|
4 |
diffusers
|
5 |
transformers
|
6 |
accelerate
|
7 |
-
spaces
|
8 |
-
git+https://github.com/XPixelGroup/BasicSR@master
|
9 |
-
gfpgan
|
10 |
-
facexlib
|
11 |
-
realesrgan
|
12 |
-
triton
|
13 |
-
xformers
|
|
|
1 |
gradio
|
2 |
+
torch==2.5.0
|
3 |
+
torch_tensorrt==2.5.0
|
4 |
+
torchvision==0.20.0
|
5 |
diffusers
|
6 |
transformers
|
7 |
accelerate
|
8 |
+
spaces
|
|
|
|
|
|
|
|
|
|
|
|