Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
e19b349
1
Parent(s):
476e594
update app
Browse files
app.py
CHANGED
@@ -1,176 +1,127 @@
|
|
1 |
import spaces
|
2 |
import torch
|
3 |
-
import time
|
4 |
import gradio as gr
|
5 |
from PIL import Image
|
6 |
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
|
7 |
-
from typing import List
|
8 |
from functools import lru_cache
|
9 |
|
10 |
MODEL_ID = "remyxai/SpaceThinker-Qwen2.5VL-3B"
|
11 |
|
12 |
-
@spaces.GPU
|
13 |
@lru_cache(maxsize=1)
|
14 |
-
def
|
15 |
-
|
16 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"
|
17 |
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
18 |
MODEL_ID,
|
19 |
-
torch_dtype=torch.bfloat16
|
20 |
-
).to(
|
21 |
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
22 |
return model, processor
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
image = image.resize((max_width, new_height), Image.Resampling.LANCZOS)
|
37 |
-
return image
|
38 |
-
|
39 |
-
def get_latest_image(history):
|
40 |
-
for item in reversed(history):
|
41 |
-
if item["role"] == "user" and isinstance(item["content"], tuple):
|
42 |
-
return item["content"][0]
|
43 |
-
return None
|
44 |
-
|
45 |
-
def only_assistant_text(full_text: str) -> str:
|
46 |
-
if "assistant" in full_text:
|
47 |
-
parts = full_text.split("assistant", 1)
|
48 |
-
result = parts[-1].strip()
|
49 |
-
result = result.lstrip(":").strip()
|
50 |
-
return result
|
51 |
-
return full_text.strip()
|
52 |
-
|
53 |
-
def run_inference(image, prompt):
|
54 |
-
model, processor = load_model()
|
55 |
system_msg = (
|
56 |
-
"You are VL-Thinking 🤔, a helpful assistant
|
57 |
-
"
|
58 |
-
"
|
59 |
)
|
60 |
conversation = [
|
61 |
-
{
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
"role": "user",
|
67 |
-
"content": [
|
68 |
-
{"type": "image", "image": image},
|
69 |
-
{"type": "text", "text": prompt},
|
70 |
-
],
|
71 |
-
},
|
72 |
]
|
73 |
-
|
|
|
|
|
74 |
conversation, tokenize=False, add_generation_prompt=True
|
75 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
-
|
78 |
-
generated_ids = model.generate(**inputs, max_new_tokens=1024)
|
79 |
-
output_text = processor.batch_decode(generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
80 |
-
return only_assistant_text(output_text)
|
81 |
|
82 |
def add_message(history, user_input):
|
83 |
-
if
|
84 |
history = []
|
85 |
-
|
86 |
-
files = user_input.get("files", [])
|
87 |
-
text = user_input.get("text", "")
|
88 |
-
|
89 |
-
for f in files:
|
90 |
history.append({"role": "user", "content": (f,)})
|
91 |
-
|
92 |
if text:
|
93 |
history.append({"role": "user", "content": text})
|
94 |
-
|
95 |
return history, gr.MultimodalTextbox(value=None)
|
96 |
|
|
|
97 |
def inference_interface(history):
|
98 |
if not history:
|
99 |
return history, gr.MultimodalTextbox(value=None)
|
100 |
-
|
101 |
-
user_text =
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
user_idx = idx
|
108 |
-
break
|
109 |
-
|
110 |
-
if user_idx == -1:
|
111 |
return history, gr.MultimodalTextbox(value=None)
|
112 |
-
|
113 |
-
|
114 |
-
|
|
|
|
|
|
|
|
|
115 |
return history, gr.MultimodalTextbox(value=None)
|
116 |
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
history.append({"role": "assistant", "content": assistant_reply})
|
121 |
return history, gr.MultimodalTextbox(value=None)
|
122 |
|
|
|
123 |
def build_demo():
|
124 |
with gr.Blocks() as demo:
|
125 |
gr.Markdown("# SpaceThinker-Qwen2.5VL-3B Image Prompt Chatbot")
|
126 |
-
|
127 |
-
chatbot = gr.Chatbot([], type="messages", line_breaks=True)
|
128 |
-
|
129 |
chat_input = gr.MultimodalTextbox(
|
130 |
interactive=True,
|
131 |
file_types=["image"],
|
132 |
placeholder="Enter text and upload an image.",
|
133 |
show_label=True
|
134 |
)
|
135 |
-
|
136 |
-
|
137 |
-
fn=add_message,
|
138 |
-
inputs=[chatbot, chat_input],
|
139 |
-
outputs=[chatbot, chat_input]
|
140 |
)
|
141 |
-
|
142 |
-
|
143 |
-
inputs=[chatbot],
|
144 |
-
outputs=[chatbot, chat_input]
|
145 |
)
|
146 |
-
|
147 |
with gr.Row():
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
fn=add_message,
|
153 |
-
inputs=[chatbot, chat_input],
|
154 |
-
outputs=[chatbot, chat_input]
|
155 |
)
|
156 |
send_click.then(
|
157 |
-
|
158 |
-
inputs=[chatbot],
|
159 |
-
outputs=[chatbot, chat_input]
|
160 |
)
|
161 |
-
|
162 |
-
gr.Examples(
|
163 |
-
examples=[
|
164 |
-
{
|
165 |
-
"text": "Give me the height of the man in the red hat in feet.",
|
166 |
-
"files": ["./examples/warehouse_rgb.jpg"]
|
167 |
-
}
|
168 |
-
],
|
169 |
-
inputs=[chat_input],
|
170 |
-
)
|
171 |
-
|
172 |
return demo
|
173 |
|
|
|
174 |
if __name__ == "__main__":
|
175 |
demo = build_demo()
|
176 |
demo.launch(share=True)
|
|
|
1 |
import spaces
|
2 |
import torch
|
|
|
3 |
import gradio as gr
|
4 |
from PIL import Image
|
5 |
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
|
|
|
6 |
from functools import lru_cache
|
7 |
|
8 |
MODEL_ID = "remyxai/SpaceThinker-Qwen2.5VL-3B"
|
9 |
|
|
|
10 |
@lru_cache(maxsize=1)
|
11 |
+
def _load_model():
|
12 |
+
"""Load and cache the model and processor inside GPU worker."""
|
|
|
13 |
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
|
14 |
MODEL_ID,
|
15 |
+
torch_dtype=torch.bfloat16
|
16 |
+
).to("cuda")
|
17 |
processor = AutoProcessor.from_pretrained(MODEL_ID)
|
18 |
return model, processor
|
19 |
|
20 |
+
@spaces.GPU
|
21 |
+
def gpu_inference(image_path: str, prompt: str) -> str:
|
22 |
+
"""Perform inference entirely in GPU subprocess."""
|
23 |
+
model, processor = _load_model()
|
24 |
+
|
25 |
+
# Load and preprocess image
|
26 |
+
image = Image.open(image_path).convert("RGB")
|
27 |
+
if image.width > 512:
|
28 |
+
ratio = image.height / image.width
|
29 |
+
image = image.resize((512, int(512 * ratio)), Image.Resampling.LANCZOS)
|
30 |
+
|
31 |
+
# Build conversation
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
32 |
system_msg = (
|
33 |
+
"You are VL-Thinking 🤔, a helpful assistant. "
|
34 |
+
"Think through your reasoning then provide the answer. "
|
35 |
+
"Wrap reasoning in <think>...</think> and final in <answer>...</answer>."
|
36 |
)
|
37 |
conversation = [
|
38 |
+
{"role": "system", "content": [{"type": "text", "text": system_msg}]},
|
39 |
+
{"role": "user", "content": [
|
40 |
+
{"type": "image", "image": image},
|
41 |
+
{"type": "text", "text": prompt}
|
42 |
+
]}
|
|
|
|
|
|
|
|
|
|
|
|
|
43 |
]
|
44 |
+
|
45 |
+
# Tokenize, generate, decode
|
46 |
+
chat_input = processor.apply_chat_template(
|
47 |
conversation, tokenize=False, add_generation_prompt=True
|
48 |
)
|
49 |
+
inputs = processor(text=[chat_input], images=[image], return_tensors="pt").to("cuda")
|
50 |
+
output_ids = model.generate(**inputs, max_new_tokens=1024)
|
51 |
+
decoded = processor.batch_decode(
|
52 |
+
output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
|
53 |
+
)[0]
|
54 |
+
|
55 |
+
# Extract assistant portion
|
56 |
+
return decoded.split("assistant", 1)[-1].strip().lstrip(":").strip()
|
57 |
|
58 |
+
# Message handling
|
|
|
|
|
|
|
59 |
|
60 |
def add_message(history, user_input):
|
61 |
+
if history is None:
|
62 |
history = []
|
63 |
+
for f in user_input.get("files", []):
|
|
|
|
|
|
|
|
|
64 |
history.append({"role": "user", "content": (f,)})
|
65 |
+
text = user_input.get("text", "")
|
66 |
if text:
|
67 |
history.append({"role": "user", "content": text})
|
|
|
68 |
return history, gr.MultimodalTextbox(value=None)
|
69 |
|
70 |
+
|
71 |
def inference_interface(history):
|
72 |
if not history:
|
73 |
return history, gr.MultimodalTextbox(value=None)
|
74 |
+
# Last user text
|
75 |
+
user_text = next(
|
76 |
+
(m["content"] for m in reversed(history)
|
77 |
+
if m["role"] == "user" and isinstance(m["content"], str)),
|
78 |
+
None
|
79 |
+
)
|
80 |
+
if user_text is None:
|
|
|
|
|
|
|
|
|
81 |
return history, gr.MultimodalTextbox(value=None)
|
82 |
+
# Last user image
|
83 |
+
image_path = next(
|
84 |
+
(m["content"][0] for m in reversed(history)
|
85 |
+
if m["role"] == "user" and isinstance(m["content"], tuple)),
|
86 |
+
None
|
87 |
+
)
|
88 |
+
if image_path is None:
|
89 |
return history, gr.MultimodalTextbox(value=None)
|
90 |
|
91 |
+
# GPU inference
|
92 |
+
reply = gpu_inference(image_path, user_text)
|
93 |
+
history.append({"role": "assistant", "content": reply})
|
|
|
94 |
return history, gr.MultimodalTextbox(value=None)
|
95 |
|
96 |
+
|
97 |
def build_demo():
|
98 |
with gr.Blocks() as demo:
|
99 |
gr.Markdown("# SpaceThinker-Qwen2.5VL-3B Image Prompt Chatbot")
|
100 |
+
chatbot = gr.Chatbot([], type="messages", label="Conversation")
|
|
|
|
|
101 |
chat_input = gr.MultimodalTextbox(
|
102 |
interactive=True,
|
103 |
file_types=["image"],
|
104 |
placeholder="Enter text and upload an image.",
|
105 |
show_label=True
|
106 |
)
|
107 |
+
submit_evt = chat_input.submit(
|
108 |
+
add_message, [chatbot, chat_input], [chatbot, chat_input]
|
|
|
|
|
|
|
109 |
)
|
110 |
+
submit_evt.then(
|
111 |
+
inference_interface, [chatbot], [chatbot, chat_input]
|
|
|
|
|
112 |
)
|
|
|
113 |
with gr.Row():
|
114 |
+
send_btn = gr.Button("Send")
|
115 |
+
clear_btn = gr.ClearButton([chatbot, chat_input])
|
116 |
+
send_click = send_btn.click(
|
117 |
+
add_message, [chatbot, chat_input], [chatbot, chat_input]
|
|
|
|
|
|
|
118 |
)
|
119 |
send_click.then(
|
120 |
+
inference_interface, [chatbot], [chatbot, chat_input]
|
|
|
|
|
121 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
122 |
return demo
|
123 |
|
124 |
+
|
125 |
if __name__ == "__main__":
|
126 |
demo = build_demo()
|
127 |
demo.launch(share=True)
|