Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -1,14 +1,23 @@
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
import random
|
|
|
4 |
from diffusers import DiffusionPipeline
|
5 |
from transformers import pipeline
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
|
7 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
8 |
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
9 |
MAX_SEED = 2**32 - 1
|
10 |
|
11 |
-
#
|
12 |
image_models = {
|
13 |
"Stable Diffusion 1.5 (light)": "runwayml/stable-diffusion-v1-5",
|
14 |
"Stable Diffusion 2.1": "stabilityai/stable-diffusion-2-1",
|
@@ -35,46 +44,136 @@ text_models = {
|
|
35 |
"LLaMA 2 7B (heavy)": "meta-llama/Llama-2-7b-hf"
|
36 |
}
|
37 |
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
image_pipes = {}
|
40 |
text_pipes = {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
|
42 |
def generate_image(prompt, model_name, seed, randomize_seed, progress=gr.Progress(track_tqdm=True)):
|
43 |
if randomize_seed:
|
44 |
seed = random.randint(0, MAX_SEED)
|
45 |
-
generator = torch.manual_seed(seed)
|
46 |
|
47 |
-
|
|
|
|
|
|
|
|
|
|
|
48 |
if model_name not in image_pipes:
|
49 |
-
|
50 |
image_models[model_name],
|
51 |
-
torch_dtype=torch_dtype
|
52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
pipe = image_pipes[model_name]
|
54 |
|
55 |
-
progress(
|
56 |
-
result = pipe(prompt=prompt, generator=
|
|
|
|
|
57 |
|
58 |
progress(100, desc="Done.")
|
59 |
-
return
|
60 |
|
61 |
def generate_text(prompt, model_name, progress=gr.Progress(track_tqdm=True)):
|
62 |
-
|
|
|
|
|
|
|
|
|
|
|
63 |
if model_name not in text_pipes:
|
64 |
-
text_pipes[model_name] = pipeline(
|
|
|
|
|
|
|
|
|
65 |
pipe = text_pipes[model_name]
|
66 |
|
67 |
-
progress(
|
68 |
result = pipe(prompt, max_length=100, do_sample=True)[0]['generated_text']
|
|
|
|
|
69 |
progress(100, desc="Done.")
|
70 |
return result
|
71 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
# Gradio Interface
|
73 |
with gr.Blocks() as demo:
|
74 |
-
gr.Markdown("#
|
75 |
|
76 |
with gr.Tabs():
|
77 |
-
#
|
78 |
with gr.Tab("🖼️ Image Generation"):
|
79 |
img_prompt = gr.Textbox(label="Prompt")
|
80 |
img_model = gr.Dropdown(choices=list(image_models.keys()), value="Stable Diffusion 1.5 (light)", label="Image Model")
|
@@ -84,20 +183,22 @@ with gr.Blocks() as demo:
|
|
84 |
img_out = gr.Image()
|
85 |
img_btn.click(fn=generate_image, inputs=[img_prompt, img_model, img_seed, img_rand], outputs=[img_out, img_seed])
|
86 |
|
87 |
-
#
|
88 |
with gr.Tab("📝 Text Generation"):
|
89 |
txt_prompt = gr.Textbox(label="Prompt")
|
90 |
txt_model = gr.Dropdown(choices=list(text_models.keys()), value="GPT-2 (light)", label="Text Model")
|
91 |
txt_btn = gr.Button("Generate Text")
|
92 |
txt_out = gr.Textbox(label="Output Text")
|
93 |
-
txt_btn.click(fn=generate_text, inputs=[txt_prompt, txt_model], outputs=txt_out)
|
94 |
|
95 |
-
#
|
96 |
-
with gr.Tab("🎥 Video Generation
|
97 |
-
gr.Markdown("⚠️ Video generation is placeholder only. Models require special setup.")
|
98 |
vid_prompt = gr.Textbox(label="Prompt")
|
99 |
-
|
100 |
-
|
101 |
-
|
|
|
|
|
|
|
102 |
|
103 |
demo.launch(show_error=True)
|
|
|
1 |
import gradio as gr
|
2 |
import torch
|
3 |
import random
|
4 |
+
import hashlib
|
5 |
from diffusers import DiffusionPipeline
|
6 |
from transformers import pipeline
|
7 |
+
from diffusers.utils import export_to_video
|
8 |
+
|
9 |
+
# Optional: xformers optimization
|
10 |
+
try:
|
11 |
+
import xformers
|
12 |
+
has_xformers = True
|
13 |
+
except ImportError:
|
14 |
+
has_xformers = False
|
15 |
|
16 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
17 |
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
|
18 |
MAX_SEED = 2**32 - 1
|
19 |
|
20 |
+
# Model lists ordered by size
|
21 |
image_models = {
|
22 |
"Stable Diffusion 1.5 (light)": "runwayml/stable-diffusion-v1-5",
|
23 |
"Stable Diffusion 2.1": "stabilityai/stable-diffusion-2-1",
|
|
|
44 |
"LLaMA 2 7B (heavy)": "meta-llama/Llama-2-7b-hf"
|
45 |
}
|
46 |
|
47 |
+
video_models = {
|
48 |
+
"CogVideoX-2B": "THUDM/CogVideoX-2b",
|
49 |
+
"CogVideoX-5B": "THUDM/CogVideoX-5b",
|
50 |
+
"AnimateDiff-Lightning": "ByteDance/AnimateDiff-Lightning",
|
51 |
+
"ModelScope T2V": "damo-vilab/text-to-video-ms-1.7b",
|
52 |
+
"VideoCrafter2": "VideoCrafter/VideoCrafter2",
|
53 |
+
"Open-Sora-Plan-v1.2.0": "LanguageBind/Open-Sora-Plan-v1.2.0",
|
54 |
+
"LTX-Video": "Lightricks/LTX-Video",
|
55 |
+
"HunyuanVideo": "tencent/HunyuanVideo",
|
56 |
+
"Latte-1": "maxin-cn/Latte-1",
|
57 |
+
"LaVie": "Vchitect/LaVie"
|
58 |
+
}
|
59 |
+
|
60 |
+
# Caches
|
61 |
image_pipes = {}
|
62 |
text_pipes = {}
|
63 |
+
video_pipes = {}
|
64 |
+
image_cache = {}
|
65 |
+
text_cache = {}
|
66 |
+
video_cache = {}
|
67 |
+
|
68 |
+
def hash_inputs(*args):
|
69 |
+
combined = "|".join(map(str, args))
|
70 |
+
return hashlib.sha256(combined.encode()).hexdigest()
|
71 |
|
72 |
def generate_image(prompt, model_name, seed, randomize_seed, progress=gr.Progress(track_tqdm=True)):
|
73 |
if randomize_seed:
|
74 |
seed = random.randint(0, MAX_SEED)
|
|
|
75 |
|
76 |
+
key = hash_inputs(prompt, model_name, seed)
|
77 |
+
if key in image_cache:
|
78 |
+
progress(100, desc="Using cached image.")
|
79 |
+
return image_cache[key], seed
|
80 |
+
|
81 |
+
progress(10, desc="Loading model...")
|
82 |
if model_name not in image_pipes:
|
83 |
+
pipe = DiffusionPipeline.from_pretrained(
|
84 |
image_models[model_name],
|
85 |
+
torch_dtype=torch_dtype,
|
86 |
+
low_cpu_mem_usage=True
|
87 |
+
)
|
88 |
+
|
89 |
+
if torch.__version__.startswith("2"):
|
90 |
+
pipe = torch.compile(pipe)
|
91 |
+
if has_xformers and device == "cuda":
|
92 |
+
try:
|
93 |
+
pipe.enable_xformers_memory_efficient_attention()
|
94 |
+
except Exception:
|
95 |
+
pass
|
96 |
+
|
97 |
+
pipe.to(device)
|
98 |
+
image_pipes[model_name] = pipe
|
99 |
+
|
100 |
pipe = image_pipes[model_name]
|
101 |
|
102 |
+
progress(40, desc="Generating image...")
|
103 |
+
result = pipe(prompt=prompt, generator=torch.manual_seed(seed), num_inference_steps=15, width=512, height=512)
|
104 |
+
image = result.images[0]
|
105 |
+
image_cache[key] = image
|
106 |
|
107 |
progress(100, desc="Done.")
|
108 |
+
return image, seed
|
109 |
|
110 |
def generate_text(prompt, model_name, progress=gr.Progress(track_tqdm=True)):
|
111 |
+
key = hash_inputs(prompt, model_name)
|
112 |
+
if key in text_cache:
|
113 |
+
progress(100, desc="Using cached text.")
|
114 |
+
return text_cache[key]
|
115 |
+
|
116 |
+
progress(10, desc="Loading model...")
|
117 |
if model_name not in text_pipes:
|
118 |
+
text_pipes[model_name] = pipeline(
|
119 |
+
"text-generation",
|
120 |
+
model=text_models[model_name],
|
121 |
+
device=0 if device == "cuda" else -1
|
122 |
+
)
|
123 |
pipe = text_pipes[model_name]
|
124 |
|
125 |
+
progress(40, desc="Generating text...")
|
126 |
result = pipe(prompt, max_length=100, do_sample=True)[0]['generated_text']
|
127 |
+
text_cache[key] = result
|
128 |
+
|
129 |
progress(100, desc="Done.")
|
130 |
return result
|
131 |
|
132 |
+
def generate_video(prompt, model_name, seed, randomize_seed, progress=gr.Progress(track_tqdm=True)):
|
133 |
+
if randomize_seed:
|
134 |
+
seed = random.randint(0, MAX_SEED)
|
135 |
+
|
136 |
+
key = hash_inputs(prompt, model_name, seed)
|
137 |
+
if key in video_cache:
|
138 |
+
progress(100, desc="Using cached video.")
|
139 |
+
return video_cache[key], seed
|
140 |
+
|
141 |
+
progress(10, desc="Loading model...")
|
142 |
+
if model_name not in video_pipes:
|
143 |
+
pipe = DiffusionPipeline.from_pretrained(
|
144 |
+
video_models[model_name],
|
145 |
+
torch_dtype=torch_dtype,
|
146 |
+
variant="fp16"
|
147 |
+
)
|
148 |
+
|
149 |
+
if torch.__version__.startswith("2"):
|
150 |
+
pipe = torch.compile(pipe)
|
151 |
+
if has_xformers and device == "cuda":
|
152 |
+
try:
|
153 |
+
pipe.enable_xformers_memory_efficient_attention()
|
154 |
+
except Exception:
|
155 |
+
pass
|
156 |
+
|
157 |
+
pipe.to(device)
|
158 |
+
video_pipes[model_name] = pipe
|
159 |
+
|
160 |
+
pipe = video_pipes[model_name]
|
161 |
+
|
162 |
+
progress(40, desc="Generating video...")
|
163 |
+
result = pipe(prompt=prompt, generator=torch.manual_seed(seed), num_inference_steps=15)
|
164 |
+
video_frames = result.frames[0]
|
165 |
+
video_path = export_to_video(video_frames)
|
166 |
+
video_cache[key] = video_path
|
167 |
+
|
168 |
+
progress(100, desc="Done.")
|
169 |
+
return video_path, seed
|
170 |
+
|
171 |
# Gradio Interface
|
172 |
with gr.Blocks() as demo:
|
173 |
+
gr.Markdown("# ⚡ Fast Multi-Model AI Playground with Caching")
|
174 |
|
175 |
with gr.Tabs():
|
176 |
+
# Image Generation
|
177 |
with gr.Tab("🖼️ Image Generation"):
|
178 |
img_prompt = gr.Textbox(label="Prompt")
|
179 |
img_model = gr.Dropdown(choices=list(image_models.keys()), value="Stable Diffusion 1.5 (light)", label="Image Model")
|
|
|
183 |
img_out = gr.Image()
|
184 |
img_btn.click(fn=generate_image, inputs=[img_prompt, img_model, img_seed, img_rand], outputs=[img_out, img_seed])
|
185 |
|
186 |
+
# Text Generation
|
187 |
with gr.Tab("📝 Text Generation"):
|
188 |
txt_prompt = gr.Textbox(label="Prompt")
|
189 |
txt_model = gr.Dropdown(choices=list(text_models.keys()), value="GPT-2 (light)", label="Text Model")
|
190 |
txt_btn = gr.Button("Generate Text")
|
191 |
txt_out = gr.Textbox(label="Output Text")
|
192 |
+
txt_btn.click(fn=generate_text, inputs=[txt_prompt, txt_model], outputs=[txt_out])
|
193 |
|
194 |
+
# Video Generation
|
195 |
+
with gr.Tab("🎥 Video Generation"):
|
|
|
196 |
vid_prompt = gr.Textbox(label="Prompt")
|
197 |
+
vid_model = gr.Dropdown(choices=list(video_models.keys()), value="CogVideoX-2B", label="Video Model")
|
198 |
+
vid_seed = gr.Slider(0, MAX_SEED, value=42, label="Seed")
|
199 |
+
vid_rand = gr.Checkbox(label="Randomize seed", value=True)
|
200 |
+
vid_btn = gr.Button("Generate Video")
|
201 |
+
vid_out = gr.Video()
|
202 |
+
vid_btn.click(fn=generate_video, inputs=[vid_prompt, vid_model, vid_seed, vid_rand], outputs=[vid_out, vid_seed])
|
203 |
|
204 |
demo.launch(show_error=True)
|