Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -1,24 +1,19 @@
|
|
1 |
import gradio as gr
|
2 |
-
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
|
|
|
3 |
from qwen_vl_utils import process_vision_info
|
4 |
import torch
|
5 |
-
|
6 |
|
7 |
-
# 指定模型路径
|
8 |
local_path = "Fancy-MLLM/R1-OneVision-7B"
|
9 |
|
10 |
-
# 加载模型和处理器
|
11 |
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
12 |
local_path, torch_dtype="auto", device_map="cpu"
|
13 |
)
|
14 |
processor = AutoProcessor.from_pretrained(local_path)
|
15 |
|
16 |
-
|
17 |
-
|
18 |
-
if image is None:
|
19 |
-
return "Error: No image uploaded!"
|
20 |
-
|
21 |
-
# 处理输入数据
|
22 |
messages = [
|
23 |
{
|
24 |
"role": "user",
|
@@ -28,11 +23,12 @@ def generate_output(image, text):
|
|
28 |
],
|
29 |
}
|
30 |
]
|
31 |
-
|
32 |
-
#
|
33 |
text_input = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
|
|
|
|
34 |
image_inputs, video_inputs = process_vision_info(messages)
|
35 |
-
|
36 |
inputs = processor(
|
37 |
text=[text_input],
|
38 |
images=image_inputs,
|
@@ -40,38 +36,109 @@ def generate_output(image, text):
|
|
40 |
padding=True,
|
41 |
return_tensors="pt",
|
42 |
)
|
43 |
-
inputs = inputs.to(model.device)
|
44 |
|
45 |
-
|
46 |
-
|
47 |
**inputs,
|
|
|
48 |
max_new_tokens=4096,
|
49 |
top_p=0.001,
|
50 |
top_k=1,
|
51 |
temperature=0.01,
|
52 |
repetition_penalty=1.0,
|
53 |
)
|
|
|
|
|
|
|
54 |
|
55 |
-
|
56 |
-
|
57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
58 |
|
59 |
# UI 组件
|
60 |
-
with gr.Blocks() as demo:
|
61 |
gr.HTML("""<center><font size=8>🦖 R1-OneVision Demo</center>""")
|
62 |
|
63 |
with gr.Row():
|
64 |
with gr.Column():
|
65 |
-
input_image = gr.Image(type="pil", label="Upload")
|
66 |
-
input_text = gr.Textbox(label="
|
67 |
with gr.Row():
|
68 |
-
|
69 |
-
|
|
|
|
|
70 |
|
71 |
with gr.Column():
|
72 |
-
output_text = gr.Markdown(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
73 |
|
74 |
-
# 绑定事件,去掉 queue=True
|
75 |
-
submit_btn.click(fn=generate_output, inputs=[input_image, input_text], outputs=output_text)
|
76 |
|
77 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor, TextIteratorStreamer
|
3 |
+
from threading import Thread
|
4 |
from qwen_vl_utils import process_vision_info
|
5 |
import torch
|
6 |
+
import time
|
7 |
|
|
|
8 |
local_path = "Fancy-MLLM/R1-OneVision-7B"
|
9 |
|
|
|
10 |
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
11 |
local_path, torch_dtype="auto", device_map="cpu"
|
12 |
)
|
13 |
processor = AutoProcessor.from_pretrained(local_path)
|
14 |
|
15 |
+
def generate_output(image, text, button_click):
|
16 |
+
# Prepare input data
|
|
|
|
|
|
|
|
|
17 |
messages = [
|
18 |
{
|
19 |
"role": "user",
|
|
|
23 |
],
|
24 |
}
|
25 |
]
|
26 |
+
|
27 |
+
# Prepare inputs for the model
|
28 |
text_input = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
29 |
+
# print(text_input)
|
30 |
+
# import pdb; pdb.set_trace()
|
31 |
image_inputs, video_inputs = process_vision_info(messages)
|
|
|
32 |
inputs = processor(
|
33 |
text=[text_input],
|
34 |
images=image_inputs,
|
|
|
36 |
padding=True,
|
37 |
return_tensors="pt",
|
38 |
)
|
39 |
+
inputs = inputs.to(model.device)
|
40 |
|
41 |
+
streamer = TextIteratorStreamer(processor, skip_prompt=True, skip_special_tokens=True)
|
42 |
+
generation_kwargs = dict(
|
43 |
**inputs,
|
44 |
+
streamer=streamer,
|
45 |
max_new_tokens=4096,
|
46 |
top_p=0.001,
|
47 |
top_k=1,
|
48 |
temperature=0.01,
|
49 |
repetition_penalty=1.0,
|
50 |
)
|
51 |
+
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
52 |
+
thread.start()
|
53 |
+
generated_text = ''
|
54 |
|
55 |
+
try:
|
56 |
+
for new_text in streamer:
|
57 |
+
generated_text += new_text
|
58 |
+
yield f"{generated_text}"
|
59 |
+
# print(f"Current text: {generated_text}") # 调试输出
|
60 |
+
# yield generated_text # 直接输出原始文本
|
61 |
+
except Exception as e:
|
62 |
+
print(f"Error: {e}")
|
63 |
+
yield f"Error occurred: {str(e)}"
|
64 |
+
|
65 |
+
Css = """
|
66 |
+
#output-markdown {
|
67 |
+
overflow-y: auto;
|
68 |
+
white-space: pre-wrap;
|
69 |
+
word-wrap: break-word;
|
70 |
+
}
|
71 |
+
|
72 |
+
#output-markdown .math {
|
73 |
+
overflow-x: auto;
|
74 |
+
max-width: 100%;
|
75 |
+
}
|
76 |
+
.markdown-text {
|
77 |
+
white-space: pre-wrap;
|
78 |
+
word-wrap: break-word;
|
79 |
+
}
|
80 |
+
#qwen-md .katex-display { display: inline; }
|
81 |
+
#qwen-md .katex-display>.katex { display: inline; }
|
82 |
+
#qwen-md .katex-display>.katex>.katex-html { display: inline; }
|
83 |
+
"""
|
84 |
|
85 |
# UI 组件
|
86 |
+
with gr.Blocks(css=Css) as demo:
|
87 |
gr.HTML("""<center><font size=8>🦖 R1-OneVision Demo</center>""")
|
88 |
|
89 |
with gr.Row():
|
90 |
with gr.Column():
|
91 |
+
input_image = gr.Image(type="pil", label="Upload"),
|
92 |
+
input_text = gr.Textbox(label="input your question")
|
93 |
with gr.Row():
|
94 |
+
with gr.Column():
|
95 |
+
clear_btn = gr.ClearButton([*input_image, input_text])
|
96 |
+
with gr.Column():
|
97 |
+
submit_btn = gr.Button("Submit", variant="primary")
|
98 |
|
99 |
with gr.Column():
|
100 |
+
output_text = gr.Markdown(
|
101 |
+
label="Generated Response",
|
102 |
+
max_height="80vh",
|
103 |
+
min_height="50vh",
|
104 |
+
container=True,
|
105 |
+
latex_delimiters=[{
|
106 |
+
"left": "\\(",
|
107 |
+
"right": "\\)",
|
108 |
+
"display": True
|
109 |
+
}, {
|
110 |
+
"left": "\\begin\{equation\}",
|
111 |
+
"right": "\\end\{equation\}",
|
112 |
+
"display": True
|
113 |
+
}, {
|
114 |
+
"left": "\\begin\{align\}",
|
115 |
+
"right": "\\end\{align\}",
|
116 |
+
"display": True
|
117 |
+
}, {
|
118 |
+
"left": "\\begin\{alignat\}",
|
119 |
+
"right": "\\end\{alignat\}",
|
120 |
+
"display": True
|
121 |
+
}, {
|
122 |
+
"left": "\\begin\{gather\}",
|
123 |
+
"right": "\\end\{gather\}",
|
124 |
+
"display": True
|
125 |
+
}, {
|
126 |
+
"left": "\\begin\{CD\}",
|
127 |
+
"right": "\\end\{CD\}",
|
128 |
+
"display": True
|
129 |
+
}, {
|
130 |
+
"left": "\\[",
|
131 |
+
"right": "\\]",
|
132 |
+
"display": True
|
133 |
+
}],
|
134 |
+
elem_id="qwen-md")
|
135 |
+
|
136 |
|
|
|
|
|
137 |
|
138 |
+
submit_btn.click(
|
139 |
+
fn=generate_output,
|
140 |
+
inputs=[*input_image, input_text],
|
141 |
+
outputs=output_text,
|
142 |
+
queue=True
|
143 |
+
)
|
144 |
+
demo.launch(share=True)
|