Spaces:
Running
on
Zero
Running
on
Zero
Upload 12 files
Browse files- .gitattributes +6 -0
- README.md +5 -4
- app.py +344 -0
- imgs/car.png +3 -0
- imgs/chair.png +3 -0
- imgs/count.png +3 -0
- imgs/foot.webp +3 -0
- imgs/table.webp +3 -0
- imgs/train.png +3 -0
- requirements.txt +29 -0
- test_img_edit.py +132 -0
- test_img_to_txt.py +84 -0
- test_txt_to_img.py +132 -0
.gitattributes
CHANGED
@@ -33,3 +33,9 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
imgs/car.png filter=lfs diff=lfs merge=lfs -text
|
37 |
+
imgs/chair.png filter=lfs diff=lfs merge=lfs -text
|
38 |
+
imgs/count.png filter=lfs diff=lfs merge=lfs -text
|
39 |
+
imgs/foot.webp filter=lfs diff=lfs merge=lfs -text
|
40 |
+
imgs/table.webp filter=lfs diff=lfs merge=lfs -text
|
41 |
+
imgs/train.png filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
@@ -1,13 +1,14 @@
|
|
1 |
---
|
2 |
title: Ovis U1 3B
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
sdk_version: 5.35.0
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
license: apache-2.0
|
|
|
11 |
---
|
12 |
|
13 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
title: Ovis U1 3B
|
3 |
+
emoji: 🎨
|
4 |
+
colorFrom: green
|
5 |
+
colorTo: indigo
|
6 |
sdk: gradio
|
7 |
sdk_version: 5.35.0
|
8 |
app_file: app.py
|
9 |
pinned: false
|
10 |
license: apache-2.0
|
11 |
+
short_description: Demo for multimodal understanding and generation
|
12 |
---
|
13 |
|
14 |
+
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
app.py
ADDED
@@ -0,0 +1,344 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import subprocess
|
3 |
+
subprocess.run('pip install flash-attn==2.6.3 --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
4 |
+
import random
|
5 |
+
import spaces
|
6 |
+
import numpy as np
|
7 |
+
import torch
|
8 |
+
from PIL import Image
|
9 |
+
import gradio as gr
|
10 |
+
from transformers import AutoModelForCausalLM
|
11 |
+
from test_img_edit import pipe_img_edit
|
12 |
+
from test_img_to_txt import pipe_txt_gen
|
13 |
+
from test_txt_to_img import pipe_t2i
|
14 |
+
|
15 |
+
|
16 |
+
# Constants
|
17 |
+
MAX_SEED = 10000
|
18 |
+
|
19 |
+
hf_token = os.getenv("HF_TOKEN")
|
20 |
+
|
21 |
+
HUB_MODEL_ID = "AIDC-AI/Ovis-U1-3B"
|
22 |
+
model, loading_info = AutoModelForCausalLM.from_pretrained(
|
23 |
+
HUB_MODEL_ID,
|
24 |
+
torch_dtype=torch.bfloat16,
|
25 |
+
output_loading_info=True,
|
26 |
+
token=hf_token,
|
27 |
+
trust_remote_code=True
|
28 |
+
)
|
29 |
+
print(f'Loading info of Ovis-U1:\n{loading_info}')
|
30 |
+
|
31 |
+
model = model.eval().to("cuda")
|
32 |
+
model = model.to(torch.bfloat16)
|
33 |
+
|
34 |
+
def set_global_seed(seed: int = 42):
|
35 |
+
random.seed(seed)
|
36 |
+
np.random.seed(seed)
|
37 |
+
torch.manual_seed(seed)
|
38 |
+
torch.cuda.manual_seed_all(seed)
|
39 |
+
|
40 |
+
def randomize_seed_fn(seed: int, randomize: bool) -> int:
|
41 |
+
return random.randint(0, MAX_SEED) if randomize else seed
|
42 |
+
|
43 |
+
@spaces.GPU
|
44 |
+
def process_txt_to_img(prompt: str, height: int, width: int, steps: int, final_seed: int, guidance_scale: float, progress: gr.Progress = gr.Progress(track_tqdm=True)) -> list[Image.Image]:
|
45 |
+
set_global_seed(final_seed)
|
46 |
+
images = pipe_t2i(model, prompt, height, width, steps, cfg=guidance_scale, seed=final_seed)
|
47 |
+
return images
|
48 |
+
|
49 |
+
@spaces.GPU
|
50 |
+
def process_img_to_txt(prompt: str, img: Image.Image, progress: gr.Progress = gr.Progress(track_tqdm=True)) -> str:
|
51 |
+
output_text = pipe_txt_gen(model, img, prompt)
|
52 |
+
return output_text
|
53 |
+
|
54 |
+
@spaces.GPU
|
55 |
+
def process_img_txt_to_img(prompt: str, img: Image.Image, steps: int, final_seed: int, txt_cfg: float, img_cfg: float, progress: gr.Progress = gr.Progress(track_tqdm=True)) -> list[Image.Image]:
|
56 |
+
set_global_seed(final_seed)
|
57 |
+
images = pipe_img_edit(model, img, prompt, steps, txt_cfg, img_cfg, seed=final_seed)
|
58 |
+
return images
|
59 |
+
|
60 |
+
# Gradio UI
|
61 |
+
with gr.Blocks(title="Ovis-U1-3B") as demo:
|
62 |
+
gr.Markdown('''# Ovis-U1-3B
|
63 |
+
''')
|
64 |
+
|
65 |
+
with gr.Row():
|
66 |
+
with gr.Column():
|
67 |
+
with gr.Tabs():
|
68 |
+
with gr.TabItem("Image + Text → Image"):
|
69 |
+
edit_image_input = gr.Image(label="Input Image", type="pil")
|
70 |
+
with gr.Row():
|
71 |
+
edit_prompt_input = gr.Textbox(
|
72 |
+
label="Prompt",
|
73 |
+
show_label=False,
|
74 |
+
placeholder="Describe the editing instruction...",
|
75 |
+
container=False,
|
76 |
+
lines=1
|
77 |
+
)
|
78 |
+
run_edit_image_btn = gr.Button("Run", scale=0)
|
79 |
+
|
80 |
+
with gr.Accordion("Advanced Settings", open=False):
|
81 |
+
|
82 |
+
with gr.Row():
|
83 |
+
|
84 |
+
edit_img_guidance_slider = gr.Slider(
|
85 |
+
label="Image Guidance Scale",
|
86 |
+
minimum=1.0, maximum=10.0,
|
87 |
+
step=0.1, value=1.5
|
88 |
+
)
|
89 |
+
|
90 |
+
edit_txt_guidance_slider = gr.Slider(
|
91 |
+
label="Text Guidance Scale",
|
92 |
+
minimum=1.0, maximum=30.0,
|
93 |
+
step=0.5, value=6.0
|
94 |
+
)
|
95 |
+
|
96 |
+
edit_num_steps_slider = gr.Slider(
|
97 |
+
label='Steps',
|
98 |
+
minimum=40, maximum=100,
|
99 |
+
value=50, step=1
|
100 |
+
)
|
101 |
+
edit_seed_slider = gr.Slider(
|
102 |
+
label="Seed",
|
103 |
+
minimum=0, maximum=int(MAX_SEED),
|
104 |
+
step=1, value=42
|
105 |
+
)
|
106 |
+
edit_randomize_checkbox = gr.Checkbox(
|
107 |
+
label="Randomize seed", value=False
|
108 |
+
)
|
109 |
+
|
110 |
+
img_edit_examples_data = [
|
111 |
+
["imgs/train.png", "Modify this image in a Ghibli style. "],
|
112 |
+
["imgs/chair.png", "Transfer the image into a faceted low-poly 3-D render style."],
|
113 |
+
["imgs/car.png", "Replace the tiny house on wheels in the image with a vintage car."],
|
114 |
+
]
|
115 |
+
gr.Examples(
|
116 |
+
examples=img_edit_examples_data,
|
117 |
+
inputs=[edit_image_input, edit_prompt_input],
|
118 |
+
cache_examples=False,
|
119 |
+
label="Image Editing Examples"
|
120 |
+
)
|
121 |
+
|
122 |
+
with gr.TabItem("Text → Image"):
|
123 |
+
with gr.Row():
|
124 |
+
prompt_gen_input = gr.Textbox(
|
125 |
+
label="Prompt",
|
126 |
+
show_label=False,
|
127 |
+
placeholder="Describe the image you want...",
|
128 |
+
container=False,
|
129 |
+
lines=1
|
130 |
+
)
|
131 |
+
run_image_gen_btn = gr.Button("Run", scale=0)
|
132 |
+
|
133 |
+
with gr.Accordion("Advanced Settings", open=False):
|
134 |
+
with gr.Row():
|
135 |
+
height_slider = gr.Slider(
|
136 |
+
label='height',
|
137 |
+
minimum=256, maximum=1536,
|
138 |
+
value=1024, step=32
|
139 |
+
)
|
140 |
+
width_slider = gr.Slider(
|
141 |
+
label='width',
|
142 |
+
minimum=256, maximum=1536,
|
143 |
+
value=1024, step=32
|
144 |
+
)
|
145 |
+
|
146 |
+
guidance_slider = gr.Slider(
|
147 |
+
label="Guidance Scale",
|
148 |
+
minimum=1.0, maximum=30.0,
|
149 |
+
step=0.5, value=5.0
|
150 |
+
)
|
151 |
+
|
152 |
+
num_steps_slider = gr.Slider(
|
153 |
+
label='Steps',
|
154 |
+
minimum=40, maximum=100,
|
155 |
+
value=50, step=1
|
156 |
+
)
|
157 |
+
seed_slider = gr.Slider(
|
158 |
+
label="Seed",
|
159 |
+
minimum=0, maximum=int(MAX_SEED),
|
160 |
+
step=1, value=42
|
161 |
+
)
|
162 |
+
randomize_checkbox = gr.Checkbox(
|
163 |
+
label="Randomize seed", value=False
|
164 |
+
)
|
165 |
+
|
166 |
+
text_gen_examples_data = [
|
167 |
+
["A breathtaking fairy with teal wings sits gracefully on a lotus flower in a serene pond, exuding elegance."],
|
168 |
+
["A winter mountain landscape at deep night with snowy terrain and colorful flowers, under beautiful clouds and no people, portrayed as an anime background illustration with intricate detail and sharp focus."],
|
169 |
+
["A photo of a pug wearing a cowboy hat and bandana, sitting on a hay bale."]
|
170 |
+
]
|
171 |
+
gr.Examples(
|
172 |
+
examples=text_gen_examples_data,
|
173 |
+
inputs=[prompt_gen_input],
|
174 |
+
cache_examples=False,
|
175 |
+
label="Image Generation Examples"
|
176 |
+
)
|
177 |
+
|
178 |
+
with gr.TabItem("Image → Text"):
|
179 |
+
image_understand_input = gr.Image(label="Input Image", type="pil")
|
180 |
+
with gr.Row():
|
181 |
+
prompt_understand_input = gr.Textbox(
|
182 |
+
label="Prompt",
|
183 |
+
show_label=False,
|
184 |
+
placeholder="Describe the question about image...",
|
185 |
+
container=False,
|
186 |
+
lines=1
|
187 |
+
)
|
188 |
+
run_image_understand_btn = gr.Button("Run", scale=0)
|
189 |
+
|
190 |
+
image_understanding_examples_data = [
|
191 |
+
["imgs/table.webp", "In what scenario does this picture take place?"],
|
192 |
+
["imgs/count.png", "How many broccoli are there in the picture?"],
|
193 |
+
["imgs/foot.webp", "Where is this picture located?"],
|
194 |
+
]
|
195 |
+
gr.Examples(
|
196 |
+
examples=image_understanding_examples_data,
|
197 |
+
inputs=[image_understand_input, prompt_understand_input],
|
198 |
+
cache_examples=False,
|
199 |
+
label="Image Understanding Examples"
|
200 |
+
)
|
201 |
+
|
202 |
+
clean_btn = gr.Button("Clear All Inputs/Outputs")
|
203 |
+
|
204 |
+
with gr.Column():
|
205 |
+
output_gallery = gr.Gallery(label="Generated Images", columns=2, visible=True) # Default to visible, content will control
|
206 |
+
output_text = gr.Textbox(label="Generated Text", visible=False, lines=5, interactive=False)
|
207 |
+
|
208 |
+
@spaces.GPU
|
209 |
+
def run_img_txt_to_img_tab(prompt, img, steps, seed, txt_cfg, img_cfg, progress=gr.Progress(track_tqdm=True)):
|
210 |
+
if img is None:
|
211 |
+
return (
|
212 |
+
gr.update(value=[], visible=False),
|
213 |
+
gr.update(value="Please upload an image for editing.", visible=True)
|
214 |
+
)
|
215 |
+
# Seed is already finalized by the randomize_seed_fn in the click chain
|
216 |
+
imgs = process_img_txt_to_img(prompt, img, steps, seed, txt_cfg, img_cfg, progress=progress)
|
217 |
+
return (
|
218 |
+
gr.update(value=imgs, visible=True),
|
219 |
+
gr.update(value="", visible=False)
|
220 |
+
)
|
221 |
+
|
222 |
+
@spaces.GPU
|
223 |
+
def run_txt_to_img_tab(prompt, height, width, steps, seed, guidance, progress=gr.Progress(track_tqdm=True)):
|
224 |
+
# Seed is already finalized by the randomize_seed_fn in the click chain
|
225 |
+
imgs = process_txt_to_img(prompt, height, width, steps, seed, guidance, progress=progress)
|
226 |
+
return (
|
227 |
+
gr.update(value=imgs, visible=True),
|
228 |
+
gr.update(value="", visible=False)
|
229 |
+
)
|
230 |
+
|
231 |
+
@spaces.GPU
|
232 |
+
def run_img_to_txt_tab(img, prompt, progress=gr.Progress(track_tqdm=True)):
|
233 |
+
if img is None:
|
234 |
+
return (
|
235 |
+
gr.update(value=[], visible=False),
|
236 |
+
gr.update(value="Please upload an image for understanding.", visible=True)
|
237 |
+
)
|
238 |
+
txt = process_img_to_txt(prompt, img, progress=progress)
|
239 |
+
return (
|
240 |
+
gr.update(value=[], visible=False),
|
241 |
+
gr.update(value=txt, visible=True)
|
242 |
+
)
|
243 |
+
|
244 |
+
def clean_all_fn():
|
245 |
+
return (
|
246 |
+
# Tab 1 inputs
|
247 |
+
gr.update(value=None),
|
248 |
+
gr.update(value=""),
|
249 |
+
gr.update(value=1.5),
|
250 |
+
gr.update(value=6.0),
|
251 |
+
gr.update(value=50),
|
252 |
+
gr.update(value=42),
|
253 |
+
gr.update(value=False),
|
254 |
+
# Tab 2 inputs
|
255 |
+
gr.update(value=""), # prompt_gen_input
|
256 |
+
gr.update(value=1024),
|
257 |
+
gr.update(value=1024),
|
258 |
+
gr.update(value=5.0),
|
259 |
+
gr.update(value=50),
|
260 |
+
gr.update(value=42), # seed_slider
|
261 |
+
gr.update(value=False), # randomize_checkbox
|
262 |
+
# Tab 3 inputs
|
263 |
+
gr.update(value=None), # image_understand_input
|
264 |
+
gr.update(value=""), # prompt_understand_input
|
265 |
+
# Outputs
|
266 |
+
gr.update(value=[], visible=True), # output_gallery (reset and keep visible for next gen)
|
267 |
+
gr.update(value="", visible=False) # output_text (reset and hide)
|
268 |
+
)
|
269 |
+
|
270 |
+
# Event listeners for Image + Text -> Image
|
271 |
+
edit_inputs = [edit_prompt_input, edit_image_input, edit_num_steps_slider, edit_seed_slider, edit_txt_guidance_slider, edit_img_guidance_slider]
|
272 |
+
|
273 |
+
run_edit_image_btn.click(
|
274 |
+
fn=randomize_seed_fn,
|
275 |
+
inputs=[edit_seed_slider, edit_randomize_checkbox],
|
276 |
+
outputs=[edit_seed_slider]
|
277 |
+
).then(
|
278 |
+
fn=run_img_txt_to_img_tab,
|
279 |
+
inputs=edit_inputs,
|
280 |
+
outputs=[output_gallery, output_text]
|
281 |
+
)
|
282 |
+
|
283 |
+
edit_prompt_input.submit(
|
284 |
+
fn=randomize_seed_fn,
|
285 |
+
inputs=[edit_seed_slider, edit_randomize_checkbox],
|
286 |
+
outputs=[edit_seed_slider]
|
287 |
+
).then(
|
288 |
+
fn=run_img_txt_to_img_tab,
|
289 |
+
inputs=edit_inputs,
|
290 |
+
outputs=[output_gallery, output_text]
|
291 |
+
)
|
292 |
+
|
293 |
+
# Event listeners for Text -> Image
|
294 |
+
gen_inputs = [prompt_gen_input, height_slider, width_slider, num_steps_slider, seed_slider, guidance_slider]
|
295 |
+
|
296 |
+
run_image_gen_btn.click(
|
297 |
+
fn=randomize_seed_fn,
|
298 |
+
inputs=[seed_slider, randomize_checkbox],
|
299 |
+
outputs=[seed_slider]
|
300 |
+
).then(
|
301 |
+
fn=run_txt_to_img_tab,
|
302 |
+
inputs=gen_inputs,
|
303 |
+
outputs=[output_gallery, output_text]
|
304 |
+
)
|
305 |
+
|
306 |
+
prompt_gen_input.submit(
|
307 |
+
fn=randomize_seed_fn,
|
308 |
+
inputs=[seed_slider, randomize_checkbox],
|
309 |
+
outputs=[seed_slider]
|
310 |
+
).then(
|
311 |
+
fn=run_txt_to_img_tab,
|
312 |
+
inputs=gen_inputs,
|
313 |
+
outputs=[output_gallery, output_text]
|
314 |
+
)
|
315 |
+
|
316 |
+
# Event listeners for Image -> Text
|
317 |
+
understand_inputs = [image_understand_input, prompt_understand_input]
|
318 |
+
|
319 |
+
run_image_understand_btn.click(
|
320 |
+
fn=run_img_to_txt_tab,
|
321 |
+
inputs=understand_inputs,
|
322 |
+
outputs=[output_gallery, output_text]
|
323 |
+
)
|
324 |
+
|
325 |
+
prompt_understand_input.submit(
|
326 |
+
fn=run_img_to_txt_tab,
|
327 |
+
inputs=understand_inputs,
|
328 |
+
outputs=[output_gallery, output_text]
|
329 |
+
)
|
330 |
+
|
331 |
+
clean_btn.click(
|
332 |
+
fn=clean_all_fn,
|
333 |
+
inputs=[],
|
334 |
+
outputs=[
|
335 |
+
edit_image_input, edit_prompt_input, edit_img_guidance_slider, edit_txt_guidance_slider,
|
336 |
+
edit_num_steps_slider, edit_seed_slider, edit_randomize_checkbox,
|
337 |
+
prompt_gen_input, height_slider, width_slider, guidance_slider, num_steps_slider, seed_slider, randomize_checkbox,
|
338 |
+
image_understand_input, prompt_understand_input,
|
339 |
+
output_gallery, output_text
|
340 |
+
]
|
341 |
+
)
|
342 |
+
|
343 |
+
if __name__ == "__main__":
|
344 |
+
demo.launch(share=True)
|
imgs/car.png
ADDED
![]() |
Git LFS Details
|
imgs/chair.png
ADDED
![]() |
Git LFS Details
|
imgs/count.png
ADDED
![]() |
Git LFS Details
|
imgs/foot.webp
ADDED
![]() |
Git LFS Details
|
imgs/table.webp
ADDED
![]() |
Git LFS Details
|
imgs/train.png
ADDED
![]() |
Git LFS Details
|
requirements.txt
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch==2.4.0
|
2 |
+
transformers==4.51.3
|
3 |
+
tokenizers==0.21.1
|
4 |
+
sentencepiece==0.1.99
|
5 |
+
pyarrow==18.0.0
|
6 |
+
accelerate==1.1.0
|
7 |
+
pydantic==2.8.2
|
8 |
+
markdown2[all]
|
9 |
+
numpy==1.24.3
|
10 |
+
scikit-learn==1.2.2
|
11 |
+
requests
|
12 |
+
httpx
|
13 |
+
uvicorn
|
14 |
+
fastapi==0.112.4
|
15 |
+
einops==0.6.1
|
16 |
+
einops-exts==0.0.4
|
17 |
+
timm==1.0.11
|
18 |
+
tiktoken
|
19 |
+
transformers_stream_generator==0.0.4
|
20 |
+
scipy
|
21 |
+
pandas
|
22 |
+
torchaudio
|
23 |
+
xformers
|
24 |
+
pillow==10.3.0
|
25 |
+
pysubs2==1.7.2
|
26 |
+
trl==0.12.1
|
27 |
+
moviepy==1.0.3
|
28 |
+
diffusers==0.31.0
|
29 |
+
gradio
|
test_img_edit.py
ADDED
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import argparse
|
3 |
+
import math
|
4 |
+
import numpy as np
|
5 |
+
import torch
|
6 |
+
from PIL import Image
|
7 |
+
from transformers import AutoModelForCausalLM
|
8 |
+
|
9 |
+
|
10 |
+
def parse_args():
|
11 |
+
parser = argparse.ArgumentParser(description="Test Image Editing")
|
12 |
+
parser.add_argument(
|
13 |
+
"--model_path",
|
14 |
+
type=str,
|
15 |
+
default="AIDC-AI/Ovis-U1-3B",
|
16 |
+
)
|
17 |
+
parser.add_argument(
|
18 |
+
"--steps", type=int, default=50,
|
19 |
+
)
|
20 |
+
parser.add_argument(
|
21 |
+
"--img_cfg", type=float, default=1.5,
|
22 |
+
)
|
23 |
+
parser.add_argument(
|
24 |
+
"--txt_cfg", type=float, default=6,
|
25 |
+
)
|
26 |
+
args = parser.parse_args()
|
27 |
+
return args
|
28 |
+
|
29 |
+
def load_blank_image(width, height):
|
30 |
+
pil_image = Image.new("RGB", (width, height), (255, 255, 255)).convert('RGB')
|
31 |
+
return pil_image
|
32 |
+
|
33 |
+
def build_inputs(model, text_tokenizer, visual_tokenizer, prompt, pil_image, target_width, target_height):
|
34 |
+
if pil_image is not None:
|
35 |
+
target_size = (int(target_width), int(target_height))
|
36 |
+
pil_image, vae_pixel_values, cond_img_ids = model.visual_generator.process_image_aspectratio(pil_image, target_size)
|
37 |
+
cond_img_ids[..., 0] = 1.0
|
38 |
+
vae_pixel_values = vae_pixel_values.unsqueeze(0).to(device=model.device)
|
39 |
+
width = pil_image.width
|
40 |
+
height = pil_image.height
|
41 |
+
resized_height, resized_width = visual_tokenizer.smart_resize(height, width, max_pixels=visual_tokenizer.image_processor.min_pixels)
|
42 |
+
pil_image = pil_image.resize((resized_width, resized_height))
|
43 |
+
else:
|
44 |
+
vae_pixel_values = None
|
45 |
+
cond_img_ids = None
|
46 |
+
|
47 |
+
prompt, input_ids, pixel_values, grid_thws = model.preprocess_inputs(
|
48 |
+
prompt,
|
49 |
+
[pil_image],
|
50 |
+
generation_preface=None,
|
51 |
+
return_labels=False,
|
52 |
+
propagate_exception=False,
|
53 |
+
multimodal_type='single_image',
|
54 |
+
fix_sample_overall_length_navit=False
|
55 |
+
)
|
56 |
+
attention_mask = torch.ne(input_ids, text_tokenizer.pad_token_id)
|
57 |
+
input_ids = input_ids.unsqueeze(0).to(device=model.device)
|
58 |
+
attention_mask = attention_mask.unsqueeze(0).to(device=model.device)
|
59 |
+
if pixel_values is not None:
|
60 |
+
pixel_values = torch.cat([
|
61 |
+
pixel_values.to(device=visual_tokenizer.device, dtype=torch.bfloat16) if pixel_values is not None else None
|
62 |
+
],dim=0)
|
63 |
+
if grid_thws is not None:
|
64 |
+
grid_thws = torch.cat([
|
65 |
+
grid_thws.to(device=visual_tokenizer.device) if grid_thws is not None else None
|
66 |
+
],dim=0)
|
67 |
+
return input_ids, pixel_values, attention_mask, grid_thws, vae_pixel_values
|
68 |
+
|
69 |
+
def pipe_img_edit(model, input_img, prompt, steps, txt_cfg, img_cfg, seed=42):
|
70 |
+
text_tokenizer = model.get_text_tokenizer()
|
71 |
+
visual_tokenizer = model.get_visual_tokenizer()
|
72 |
+
|
73 |
+
width, height = input_img.size
|
74 |
+
height, width = visual_tokenizer.smart_resize(height, width, factor=32)
|
75 |
+
|
76 |
+
gen_kwargs = dict(
|
77 |
+
max_new_tokens=1024,
|
78 |
+
do_sample=False,
|
79 |
+
top_p=None,
|
80 |
+
top_k=None,
|
81 |
+
temperature=None,
|
82 |
+
repetition_penalty=None,
|
83 |
+
eos_token_id=text_tokenizer.eos_token_id,
|
84 |
+
pad_token_id=text_tokenizer.pad_token_id,
|
85 |
+
use_cache=True,
|
86 |
+
height=height,
|
87 |
+
width=width,
|
88 |
+
num_steps=steps,
|
89 |
+
seed=seed,
|
90 |
+
img_cfg=img_cfg,
|
91 |
+
txt_cfg=txt_cfg,
|
92 |
+
)
|
93 |
+
uncond_image = load_blank_image(width, height)
|
94 |
+
uncond_prompt = "<image>\nGenerate an image."
|
95 |
+
input_ids, pixel_values, attention_mask, grid_thws, _ = build_inputs(model, text_tokenizer, visual_tokenizer, uncond_prompt, uncond_image, width, height)
|
96 |
+
with torch.inference_mode():
|
97 |
+
no_both_cond = model.generate_condition(input_ids, pixel_values=pixel_values, attention_mask=attention_mask, grid_thws=grid_thws, **gen_kwargs)
|
98 |
+
|
99 |
+
input_img = input_img.resize((width, height))
|
100 |
+
prompt = "<image>\n" + prompt.strip()
|
101 |
+
with torch.inference_mode():
|
102 |
+
input_ids, pixel_values, attention_mask, grid_thws, _ = build_inputs(model, text_tokenizer, visual_tokenizer, uncond_prompt, input_img, width, height)
|
103 |
+
no_txt_cond = model.generate_condition(input_ids, pixel_values=pixel_values, attention_mask=attention_mask, grid_thws=grid_thws, **gen_kwargs)
|
104 |
+
|
105 |
+
input_ids, pixel_values, attention_mask, grid_thws, vae_pixel_values = build_inputs(model, text_tokenizer, visual_tokenizer, prompt, input_img, width, height)
|
106 |
+
with torch.inference_mode():
|
107 |
+
cond = model.generate_condition(input_ids, pixel_values=pixel_values, attention_mask=attention_mask, grid_thws=grid_thws, **gen_kwargs)
|
108 |
+
cond["vae_pixel_values"] = vae_pixel_values
|
109 |
+
images = model.generate_img(cond=cond, no_both_cond=no_both_cond, no_txt_cond=no_txt_cond, **gen_kwargs)
|
110 |
+
return images
|
111 |
+
|
112 |
+
def main():
|
113 |
+
args = parse_args()
|
114 |
+
model, loading_info = AutoModelForCausalLM.from_pretrained(args.model_path,
|
115 |
+
torch_dtype=torch.bfloat16,
|
116 |
+
output_loading_info=True,
|
117 |
+
trust_remote_code=True
|
118 |
+
)
|
119 |
+
print(f'Loading info of Ovis-U1:\n{loading_info}')
|
120 |
+
|
121 |
+
model = model.eval().to("cuda")
|
122 |
+
model = model.to(torch.bfloat16)
|
123 |
+
image_path = os.path.join(os.path.dirname(__file__), "docs", "imgs", "cat.png")
|
124 |
+
pil_img = Image.open(image_path).convert('RGB')
|
125 |
+
prompt = "add a hat to this cat."
|
126 |
+
image = pipe_img_edit(model, pil_img, prompt,
|
127 |
+
args.steps, args.txt_cfg, args.img_cfg)[0]
|
128 |
+
image.save("test_image_edit.png")
|
129 |
+
|
130 |
+
|
131 |
+
if __name__ == "__main__":
|
132 |
+
main()
|
test_img_to_txt.py
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import argparse
|
3 |
+
import torch
|
4 |
+
from PIL import Image
|
5 |
+
from transformers import AutoModelForCausalLM
|
6 |
+
|
7 |
+
def parse_args():
|
8 |
+
parser = argparse.ArgumentParser(description="Test Text Generation")
|
9 |
+
parser.add_argument(
|
10 |
+
"--model_path",
|
11 |
+
type=str,
|
12 |
+
default="AIDC-AI/Ovis-U1-3B",
|
13 |
+
)
|
14 |
+
args = parser.parse_args()
|
15 |
+
return args
|
16 |
+
|
17 |
+
|
18 |
+
def build_inputs(model, text_tokenizer, visual_tokenizer, prompt, pil_image):
|
19 |
+
prompt, input_ids, pixel_values, grid_thws = model.preprocess_inputs(
|
20 |
+
prompt,
|
21 |
+
[pil_image],
|
22 |
+
generation_preface=None,
|
23 |
+
return_labels=False,
|
24 |
+
propagate_exception=False,
|
25 |
+
multimodal_type='single_image',
|
26 |
+
fix_sample_overall_length_navit=False
|
27 |
+
)
|
28 |
+
attention_mask = torch.ne(input_ids, text_tokenizer.pad_token_id)
|
29 |
+
input_ids = input_ids.unsqueeze(0).to(device=model.device)
|
30 |
+
attention_mask = attention_mask.unsqueeze(0).to(device=model.device)
|
31 |
+
if pixel_values is not None:
|
32 |
+
pixel_values = torch.cat([
|
33 |
+
pixel_values.to(device=visual_tokenizer.device, dtype=torch.bfloat16) if pixel_values is not None else None
|
34 |
+
],dim=0)
|
35 |
+
if grid_thws is not None:
|
36 |
+
grid_thws = torch.cat([
|
37 |
+
grid_thws.to(device=visual_tokenizer.device) if grid_thws is not None else None
|
38 |
+
],dim=0)
|
39 |
+
return input_ids, pixel_values, attention_mask, grid_thws
|
40 |
+
|
41 |
+
|
42 |
+
def pipe_txt_gen(model, pil_image, prompt):
|
43 |
+
text_tokenizer = model.get_text_tokenizer()
|
44 |
+
visual_tokenizer = model.get_visual_tokenizer()
|
45 |
+
gen_kwargs = dict(
|
46 |
+
max_new_tokens=4096,
|
47 |
+
do_sample=False,
|
48 |
+
top_p=None,
|
49 |
+
top_k=None,
|
50 |
+
temperature=None,
|
51 |
+
repetition_penalty=None,
|
52 |
+
eos_token_id=text_tokenizer.eos_token_id,
|
53 |
+
pad_token_id=text_tokenizer.pad_token_id,
|
54 |
+
use_cache=True,
|
55 |
+
)
|
56 |
+
prompt = "<image>\n" + prompt
|
57 |
+
input_ids, pixel_values, attention_mask, grid_thws = build_inputs(model, text_tokenizer, visual_tokenizer, prompt, pil_image)
|
58 |
+
with torch.inference_mode():
|
59 |
+
output_ids = model.generate(input_ids, pixel_values=pixel_values, attention_mask=attention_mask, grid_thws=grid_thws, **gen_kwargs)[0]
|
60 |
+
gen_text = text_tokenizer.decode(output_ids, skip_special_tokens=True)
|
61 |
+
return gen_text
|
62 |
+
|
63 |
+
|
64 |
+
def main():
|
65 |
+
# load model
|
66 |
+
args = parse_args()
|
67 |
+
model, loading_info = AutoModelForCausalLM.from_pretrained(args.model_path,
|
68 |
+
torch_dtype=torch.bfloat16,
|
69 |
+
output_loading_info=True,
|
70 |
+
trust_remote_code=True
|
71 |
+
)
|
72 |
+
print(f'Loading info of Ovis-U1:\n{loading_info}')
|
73 |
+
|
74 |
+
model = model.eval().to("cuda")
|
75 |
+
model = model.to(torch.bfloat16)
|
76 |
+
image_path = os.path.join(os.path.dirname(__file__), "docs", "imgs", "cat.png")
|
77 |
+
pil_img = Image.open(image_path).convert('RGB')
|
78 |
+
prompt = "What is it?"
|
79 |
+
gen_txt = pipe_txt_gen(model, pil_img, prompt)
|
80 |
+
print(gen_txt)
|
81 |
+
|
82 |
+
|
83 |
+
if __name__ == "__main__":
|
84 |
+
main()
|
test_txt_to_img.py
ADDED
@@ -0,0 +1,132 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import argparse
|
3 |
+
import math
|
4 |
+
import torch
|
5 |
+
from PIL import Image
|
6 |
+
from transformers import AutoModelForCausalLM
|
7 |
+
|
8 |
+
|
9 |
+
def parse_args():
|
10 |
+
parser = argparse.ArgumentParser(description="Test Text-to-Image")
|
11 |
+
parser.add_argument(
|
12 |
+
"--model_path",
|
13 |
+
type=str,
|
14 |
+
default="AIDC-AI/Ovis-U1-3B",
|
15 |
+
)
|
16 |
+
parser.add_argument(
|
17 |
+
"--height",
|
18 |
+
type=int,
|
19 |
+
default=1024,
|
20 |
+
)
|
21 |
+
parser.add_argument(
|
22 |
+
"--width",
|
23 |
+
type=int,
|
24 |
+
default=1024,
|
25 |
+
)
|
26 |
+
parser.add_argument(
|
27 |
+
"--seed", type=int, default=42,
|
28 |
+
)
|
29 |
+
parser.add_argument(
|
30 |
+
"--steps", type=int, default=50,
|
31 |
+
)
|
32 |
+
parser.add_argument(
|
33 |
+
"--txt_cfg", type=float, default=5,
|
34 |
+
)
|
35 |
+
args = parser.parse_args()
|
36 |
+
return args
|
37 |
+
|
38 |
+
|
39 |
+
def load_blank_image(width, height):
|
40 |
+
pil_image = Image.new("RGB", (width, height), (255, 255, 255)).convert('RGB')
|
41 |
+
return pil_image
|
42 |
+
|
43 |
+
def build_inputs(model, text_tokenizer, visual_tokenizer, prompt, pil_image, target_width, target_height):
|
44 |
+
if pil_image is not None:
|
45 |
+
target_size = (int(target_width), int(target_height))
|
46 |
+
pil_image, vae_pixel_values, cond_img_ids = model.visual_generator.process_image_aspectratio(pil_image, target_size)
|
47 |
+
cond_img_ids[..., 0] = 1.0
|
48 |
+
vae_pixel_values = vae_pixel_values.unsqueeze(0).to(device=model.device)
|
49 |
+
width = pil_image.width
|
50 |
+
height = pil_image.height
|
51 |
+
resized_height, resized_width = visual_tokenizer.smart_resize(height, width, max_pixels=visual_tokenizer.image_processor.min_pixels)
|
52 |
+
pil_image = pil_image.resize((resized_width, resized_height))
|
53 |
+
else:
|
54 |
+
vae_pixel_values = None
|
55 |
+
cond_img_ids = None
|
56 |
+
|
57 |
+
prompt, input_ids, pixel_values, grid_thws = model.preprocess_inputs(
|
58 |
+
prompt,
|
59 |
+
[pil_image],
|
60 |
+
generation_preface=None,
|
61 |
+
return_labels=False,
|
62 |
+
propagate_exception=False,
|
63 |
+
multimodal_type='single_image',
|
64 |
+
fix_sample_overall_length_navit=False
|
65 |
+
)
|
66 |
+
attention_mask = torch.ne(input_ids, text_tokenizer.pad_token_id)
|
67 |
+
input_ids = input_ids.unsqueeze(0).to(device=model.device)
|
68 |
+
attention_mask = attention_mask.unsqueeze(0).to(device=model.device)
|
69 |
+
if pixel_values is not None:
|
70 |
+
pixel_values = torch.cat([
|
71 |
+
pixel_values.to(device=visual_tokenizer.device, dtype=torch.bfloat16) if pixel_values is not None else None
|
72 |
+
],dim=0)
|
73 |
+
if grid_thws is not None:
|
74 |
+
grid_thws = torch.cat([
|
75 |
+
grid_thws.to(device=visual_tokenizer.device) if grid_thws is not None else None
|
76 |
+
],dim=0)
|
77 |
+
return input_ids, pixel_values, attention_mask, grid_thws, vae_pixel_values
|
78 |
+
|
79 |
+
|
80 |
+
def pipe_t2i(model, prompt, height, width, steps, cfg, seed=42):
|
81 |
+
text_tokenizer = model.get_text_tokenizer()
|
82 |
+
visual_tokenizer = model.get_visual_tokenizer()
|
83 |
+
gen_kwargs = dict(
|
84 |
+
max_new_tokens=1024,
|
85 |
+
do_sample=False,
|
86 |
+
top_p=None,
|
87 |
+
top_k=None,
|
88 |
+
temperature=None,
|
89 |
+
repetition_penalty=None,
|
90 |
+
eos_token_id=text_tokenizer.eos_token_id,
|
91 |
+
pad_token_id=text_tokenizer.pad_token_id,
|
92 |
+
use_cache=True,
|
93 |
+
height=height,
|
94 |
+
width=width,
|
95 |
+
num_steps=steps,
|
96 |
+
seed=seed,
|
97 |
+
img_cfg=0,
|
98 |
+
txt_cfg=cfg,
|
99 |
+
)
|
100 |
+
uncond_image = load_blank_image(width, height)
|
101 |
+
uncond_prompt = "<image>\nGenerate an image."
|
102 |
+
input_ids, pixel_values, attention_mask, grid_thws, _ = build_inputs(model, text_tokenizer, visual_tokenizer, uncond_prompt, uncond_image, width, height)
|
103 |
+
with torch.inference_mode():
|
104 |
+
no_both_cond = model.generate_condition(input_ids, pixel_values=pixel_values, attention_mask=attention_mask, grid_thws=grid_thws, **gen_kwargs)
|
105 |
+
prompt = "<image>\nDescribe the image by detailing the color, shape, size, texture, quantity, text, and spatial relationships of the objects:" + prompt
|
106 |
+
no_txt_cond = None
|
107 |
+
input_ids, pixel_values, attention_mask, grid_thws, vae_pixel_values = build_inputs(model, text_tokenizer, visual_tokenizer, prompt, uncond_image, width, height)
|
108 |
+
with torch.inference_mode():
|
109 |
+
cond = model.generate_condition(input_ids, pixel_values=pixel_values, attention_mask=attention_mask, grid_thws=grid_thws, **gen_kwargs)
|
110 |
+
cond["vae_pixel_values"] = vae_pixel_values
|
111 |
+
images = model.generate_img(cond=cond, no_both_cond=no_both_cond, no_txt_cond=no_txt_cond, **gen_kwargs)
|
112 |
+
return images
|
113 |
+
|
114 |
+
|
115 |
+
def main():
|
116 |
+
args = parse_args()
|
117 |
+
model, loading_info = AutoModelForCausalLM.from_pretrained(args.model_path,
|
118 |
+
torch_dtype=torch.bfloat16,
|
119 |
+
output_loading_info=True,
|
120 |
+
trust_remote_code=True
|
121 |
+
)
|
122 |
+
print(f'Loading info of Ovis-U1:\n{loading_info}')
|
123 |
+
|
124 |
+
model = model.eval().to("cuda")
|
125 |
+
model = model.to(torch.bfloat16)
|
126 |
+
prompt = "a cute cat"
|
127 |
+
image = pipe_t2i(model, prompt, args.height, args.width, args.steps, args.txt_cfg)[0]
|
128 |
+
image.save("test_t2i.png")
|
129 |
+
|
130 |
+
|
131 |
+
if __name__ == "__main__":
|
132 |
+
main()
|