Update app.py
Browse files
app.py
CHANGED
@@ -1,37 +1,53 @@
|
|
1 |
-
|
2 |
-
import spaces
|
3 |
-
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor, TextIteratorStreamer
|
4 |
-
from qwen_vl_utils import process_vision_info
|
5 |
-
import torch
|
6 |
-
from PIL import Image
|
7 |
-
import subprocess
|
8 |
-
import numpy as np
|
9 |
import os
|
10 |
-
from
|
|
|
|
|
11 |
import uuid
|
12 |
import io
|
|
|
13 |
|
14 |
-
#
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
|
|
|
|
|
|
25 |
image_extensions = Image.registered_extensions()
|
26 |
-
video_extensions = ("avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg", "
|
27 |
|
28 |
|
29 |
def identify_and_save_blob(blob_path):
|
30 |
-
"""
|
|
|
|
|
|
|
31 |
try:
|
32 |
with open(blob_path, 'rb') as file:
|
33 |
blob_content = file.read()
|
34 |
-
|
35 |
# Try to identify if it's an image
|
36 |
try:
|
37 |
Image.open(io.BytesIO(blob_content)).verify() # Check if it's a valid image
|
@@ -39,106 +55,184 @@ def identify_and_save_blob(blob_path):
|
|
39 |
media_type = "image"
|
40 |
except (IOError, SyntaxError):
|
41 |
# If it's not a valid image, assume it's a video
|
42 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
media_type = "video"
|
44 |
-
|
45 |
# Create a unique filename
|
46 |
filename = f"temp_{uuid.uuid4()}_media{extension}"
|
47 |
with open(filename, "wb") as f:
|
48 |
f.write(blob_content)
|
49 |
-
|
50 |
return filename, media_type
|
51 |
-
|
52 |
except FileNotFoundError:
|
53 |
raise ValueError(f"The file {blob_path} was not found.")
|
54 |
except Exception as e:
|
55 |
raise ValueError(f"An error occurred while processing the file: {e}")
|
56 |
|
57 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
@spaces.GPU
|
59 |
-
def
|
60 |
-
if
|
61 |
-
|
62 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
63 |
media_type = "image"
|
64 |
-
elif
|
|
|
65 |
media_type = "video"
|
66 |
else:
|
|
|
67 |
try:
|
68 |
-
media_path, media_type = identify_and_save_blob(
|
69 |
-
print(
|
70 |
except Exception as e:
|
71 |
-
print(e)
|
72 |
-
raise
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
|
78 |
-
|
79 |
-
|
80 |
-
{
|
81 |
-
"role": "user",
|
82 |
-
"content": [
|
83 |
-
{
|
84 |
-
"type": media_type,
|
85 |
-
media_type: media_path,
|
86 |
-
**({"fps": 8.0} if media_type == "video" else {}),
|
87 |
-
},
|
88 |
-
{"type": "text", "text": text_input},
|
89 |
-
],
|
90 |
-
}
|
91 |
-
]
|
92 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
93 |
text = processor.apply_chat_template(
|
94 |
messages, tokenize=False, add_generation_prompt=True
|
95 |
)
|
96 |
-
image_inputs, video_inputs = process_vision_info(messages)
|
97 |
inputs = processor(
|
98 |
text=[text],
|
99 |
images=image_inputs,
|
100 |
videos=video_inputs,
|
101 |
padding=True,
|
102 |
return_tensors="pt",
|
103 |
-
).to(
|
104 |
|
|
|
105 |
streamer = TextIteratorStreamer(
|
106 |
processor, skip_prompt=True, **{"skip_special_tokens": True}
|
107 |
)
|
108 |
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
|
109 |
|
|
|
110 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
111 |
thread.start()
|
112 |
|
113 |
buffer = ""
|
114 |
for new_text in streamer:
|
115 |
buffer += new_text
|
116 |
-
yield buffer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
css = """
|
119 |
#output {
|
120 |
-
height: 500px;
|
121 |
-
overflow: auto;
|
122 |
-
border: 1px solid #ccc;
|
123 |
}
|
124 |
"""
|
125 |
|
126 |
with gr.Blocks(css=css) as demo:
|
127 |
gr.Markdown(DESCRIPTION)
|
128 |
-
|
129 |
-
with gr.Tab(label="Image/Video Input"):
|
130 |
with gr.Row():
|
131 |
with gr.Column():
|
|
|
132 |
input_media = gr.File(
|
133 |
-
label="Upload Image or Video
|
|
|
134 |
)
|
135 |
-
|
|
|
|
|
|
|
136 |
submit_btn = gr.Button(value="Submit")
|
137 |
with gr.Column():
|
138 |
-
output_text = gr.Textbox(label="Output Text")
|
|
|
139 |
|
140 |
-
submit_btn.click(
|
141 |
-
|
142 |
-
|
143 |
|
144 |
demo.launch(debug=True)
|
|
|
1 |
+
# Standard library imports
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
import os
|
3 |
+
from datetime import datetime
|
4 |
+
import subprocess
|
5 |
+
import time
|
6 |
import uuid
|
7 |
import io
|
8 |
+
from threading import Thread
|
9 |
|
10 |
+
# Third-party imports
|
11 |
+
import numpy as np
|
12 |
+
import torch
|
13 |
+
from PIL import Image
|
14 |
+
import accelerate
|
15 |
+
import gradio as gr
|
16 |
+
import spaces
|
17 |
+
from transformers import (
|
18 |
+
Qwen2_5_VLForConditionalGeneration,
|
19 |
+
AutoTokenizer,
|
20 |
+
AutoProcessor,
|
21 |
+
TextIteratorStreamer
|
22 |
+
)
|
23 |
+
|
24 |
+
# Local imports
|
25 |
+
from qwen_vl_utils import process_vision_info
|
26 |
|
27 |
+
# Set device agnostic code
|
28 |
+
if torch.cuda.is_available():
|
29 |
+
device = "cuda"
|
30 |
+
elif (torch.backends.mps.is_available()) and (torch.backends.mps.is_built()):
|
31 |
+
device = "mps"
|
32 |
+
else:
|
33 |
+
device = "cpu"
|
34 |
|
35 |
+
print(f"[INFO] Using device: {device}")
|
36 |
+
|
37 |
+
# Define supported media extensions
|
38 |
image_extensions = Image.registered_extensions()
|
39 |
+
video_extensions = ("avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg", "gif", "webm", "m4v", "3gp") # Removed .wav as it's audio, not video
|
40 |
|
41 |
|
42 |
def identify_and_save_blob(blob_path):
|
43 |
+
"""
|
44 |
+
Identifies if the blob is an image or video and saves it with a unique name.
|
45 |
+
Returns the saved file path and its media type ("image" or "video").
|
46 |
+
"""
|
47 |
try:
|
48 |
with open(blob_path, 'rb') as file:
|
49 |
blob_content = file.read()
|
50 |
+
|
51 |
# Try to identify if it's an image
|
52 |
try:
|
53 |
Image.open(io.BytesIO(blob_content)).verify() # Check if it's a valid image
|
|
|
55 |
media_type = "image"
|
56 |
except (IOError, SyntaxError):
|
57 |
# If it's not a valid image, assume it's a video
|
58 |
+
# We can try to get the actual extension from the blob_path,
|
59 |
+
# but for unknown types, MP4 is a good default.
|
60 |
+
_, ext = os.path.splitext(blob_path)
|
61 |
+
if ext.lower() in video_extensions:
|
62 |
+
extension = ext.lower()
|
63 |
+
else:
|
64 |
+
extension = ".mp4" # Default to MP4 for saving
|
65 |
media_type = "video"
|
66 |
+
|
67 |
# Create a unique filename
|
68 |
filename = f"temp_{uuid.uuid4()}_media{extension}"
|
69 |
with open(filename, "wb") as f:
|
70 |
f.write(blob_content)
|
71 |
+
|
72 |
return filename, media_type
|
73 |
+
|
74 |
except FileNotFoundError:
|
75 |
raise ValueError(f"The file {blob_path} was not found.")
|
76 |
except Exception as e:
|
77 |
raise ValueError(f"An error occurred while processing the file: {e}")
|
78 |
|
79 |
|
80 |
+
# Model and Processor Loading
|
81 |
+
# Define models and processors as dictionaries for easy selection
|
82 |
+
models = {
|
83 |
+
"Qwen/Qwen2.5-VL-7B-Instruct": Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct",
|
84 |
+
trust_remote_code=True,
|
85 |
+
torch_dtype="auto",
|
86 |
+
device_map="auto").eval(),
|
87 |
+
"Qwen/Qwen2.5-VL-3B-Instruct": Qwen2_5_VLForConditionalGeneration.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct",
|
88 |
+
trust_remote_code=True,
|
89 |
+
torch_dtype="auto",
|
90 |
+
device_map="auto").eval()
|
91 |
+
}
|
92 |
+
|
93 |
+
processors = {
|
94 |
+
"Qwen/Qwen2.5-VL-7B-Instruct": AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-7B-Instruct", trust_remote_code=True),
|
95 |
+
"Qwen/Qwen2.5-VL-3B-Instruct": AutoProcessor.from_pretrained("Qwen/Qwen2.5-VL-3B-Instruct", trust_remote_code=True)
|
96 |
+
}
|
97 |
+
|
98 |
+
DESCRIPTION = "[Qwen2.5-VL Demo](https://huggingface.co/collections/Qwen/qwen25-vl-6795ffac22b334a837c0f9a5)"
|
99 |
+
|
100 |
@spaces.GPU
|
101 |
+
def run_example(media_input, text_input=None, model_id=None):
|
102 |
+
if media_input is None:
|
103 |
+
raise gr.Error("No media provided. Please upload an image or video before submitting.")
|
104 |
+
if model_id is None:
|
105 |
+
raise gr.Error("No model selected. Please select a model.")
|
106 |
+
|
107 |
+
start_time = time.time()
|
108 |
+
|
109 |
+
media_path = None
|
110 |
+
media_type = None
|
111 |
+
|
112 |
+
# Determine if it's an image (numpy array from gr.Image) or a file (from gr.File)
|
113 |
+
if isinstance(media_input, np.ndarray): # This comes from gr.Image
|
114 |
+
img = Image.fromarray(np.uint8(media_input))
|
115 |
+
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
|
116 |
+
filename = f"image_{timestamp}.png"
|
117 |
+
img.save(filename)
|
118 |
+
media_path = os.path.abspath(filename)
|
119 |
+
media_type = "image"
|
120 |
+
elif isinstance(media_input, str): # This comes from gr.File (filepath)
|
121 |
+
path = media_input
|
122 |
+
_, ext = os.path.splitext(path)
|
123 |
+
ext = ext.lower()
|
124 |
+
|
125 |
+
if ext in image_extensions:
|
126 |
+
media_path = path
|
127 |
media_type = "image"
|
128 |
+
elif ext in video_extensions:
|
129 |
+
media_path = path
|
130 |
media_type = "video"
|
131 |
else:
|
132 |
+
# For blobs or unknown file types, try to identify
|
133 |
try:
|
134 |
+
media_path, media_type = identify_and_save_blob(path)
|
135 |
+
print(f"Identified blob as: {media_type}, saved to: {media_path}")
|
136 |
except Exception as e:
|
137 |
+
print(f"Error identifying blob: {e}")
|
138 |
+
raise gr.Error("Unsupported media type. Please upload an image (PNG, JPG, etc.) or a video (MP4, AVI, etc.).")
|
139 |
+
else:
|
140 |
+
raise gr.Error("Unsupported input type for media. Please upload an image or video.")
|
141 |
+
|
142 |
+
print(f"[INFO] Processing {media_type} from {media_path}")
|
143 |
+
|
144 |
+
model = models[model_id]
|
145 |
+
processor = processors[model_id]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
146 |
|
147 |
+
# Construct messages list based on media type
|
148 |
+
content_list = []
|
149 |
+
if media_type == "image":
|
150 |
+
content_list.append({"type": "image", "image": media_path})
|
151 |
+
elif media_type == "video":
|
152 |
+
content_list.append({"type": "video", "video": media_path, "fps": 8.0}) # Qwen2.5-VL often uses 8fps
|
153 |
+
|
154 |
+
if text_input:
|
155 |
+
content_list.append({"type": "text", "text": text_input})
|
156 |
+
else:
|
157 |
+
# Default prompt if no text_input is provided
|
158 |
+
content_list.append({"type": "text", "text": "What is in this image/video?"})
|
159 |
+
|
160 |
+
|
161 |
+
messages = [{"role": "user", "content": content_list}]
|
162 |
+
|
163 |
+
# Preparation for inference
|
164 |
text = processor.apply_chat_template(
|
165 |
messages, tokenize=False, add_generation_prompt=True
|
166 |
)
|
167 |
+
image_inputs, video_inputs = process_vision_info(messages) # This utility handles both image and video info
|
168 |
inputs = processor(
|
169 |
text=[text],
|
170 |
images=image_inputs,
|
171 |
videos=video_inputs,
|
172 |
padding=True,
|
173 |
return_tensors="pt",
|
174 |
+
).to(device)
|
175 |
|
176 |
+
# Inference: Generation of the output using streaming
|
177 |
streamer = TextIteratorStreamer(
|
178 |
processor, skip_prompt=True, **{"skip_special_tokens": True}
|
179 |
)
|
180 |
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
|
181 |
|
182 |
+
# Start generation in a separate thread to allow streaming
|
183 |
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
184 |
thread.start()
|
185 |
|
186 |
buffer = ""
|
187 |
for new_text in streamer:
|
188 |
buffer += new_text
|
189 |
+
yield buffer, None # Yield partial text and None for time until full generation
|
190 |
+
# Clean up the temporary file after it's processed (optional, depends on use case)
|
191 |
+
# if media_path and os.path.exists(media_path) and "temp_" in os.path.basename(media_path):
|
192 |
+
# os.remove(media_path)
|
193 |
+
|
194 |
+
|
195 |
+
end_time = time.time()
|
196 |
+
total_time = round(end_time - start_time, 2)
|
197 |
+
|
198 |
+
# Final yield with total time
|
199 |
+
yield buffer, f"{total_time} seconds"
|
200 |
+
|
201 |
+
# Clean up the temporary file after it's fully processed
|
202 |
+
if media_path and os.path.exists(media_path) and "temp_" in os.path.basename(media_path):
|
203 |
+
os.remove(media_path)
|
204 |
+
print(f"[INFO] Cleaned up temporary file: {media_path}")
|
205 |
+
|
206 |
|
207 |
css = """
|
208 |
#output {
|
209 |
+
height: 500px;
|
210 |
+
overflow: auto;
|
211 |
+
border: 1px solid #ccc;
|
212 |
}
|
213 |
"""
|
214 |
|
215 |
with gr.Blocks(css=css) as demo:
|
216 |
gr.Markdown(DESCRIPTION)
|
217 |
+
with gr.Tab(label="Qwen2.5-VL Input"):
|
|
|
218 |
with gr.Row():
|
219 |
with gr.Column():
|
220 |
+
# Change input to gr.File to accept both image and video
|
221 |
input_media = gr.File(
|
222 |
+
label="Upload Image or Video (JPG, PNG, MP4, AVI, etc.)",
|
223 |
+
type="filepath" # Use 'filepath' to get the path to the temp file
|
224 |
)
|
225 |
+
model_selector = gr.Dropdown(choices=list(models.keys()),
|
226 |
+
label="Model",
|
227 |
+
value="Qwen/Qwen2.5-VL-7B-Instruct")
|
228 |
+
text_input = gr.Textbox(label="Text Prompt")
|
229 |
submit_btn = gr.Button(value="Submit")
|
230 |
with gr.Column():
|
231 |
+
output_text = gr.Textbox(label="Output Text", interactive=False)
|
232 |
+
time_taken = gr.Textbox(label="Time taken for processing + inference", interactive=False)
|
233 |
|
234 |
+
submit_btn.click(run_example,
|
235 |
+
[input_media, text_input, model_selector],
|
236 |
+
[output_text, time_taken]) # Ensure output components match yield order
|
237 |
|
238 |
demo.launch(debug=True)
|