Spaces:
Runtime error
Runtime error
Huayang Li
commited on
Commit
·
d711bd7
1
Parent(s):
89690cf
update demo with case
Browse files- app_case.py +234 -0
app_case.py
ADDED
@@ -0,0 +1,234 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoModel, AutoTokenizer
|
2 |
+
import os
|
3 |
+
import ipdb
|
4 |
+
import gradio as gr
|
5 |
+
import mdtex2html
|
6 |
+
from model.openllama import OpenLLAMAPEFTModel
|
7 |
+
import torch
|
8 |
+
import json
|
9 |
+
from header import TaskType, LoraConfig
|
10 |
+
|
11 |
+
# init the model
|
12 |
+
args = {
|
13 |
+
'model': 'openllama_peft',
|
14 |
+
'imagebind_ckpt_path': 'pretrained_ckpt/imagebind_ckpt',
|
15 |
+
'vicuna_ckpt_path': 'openllmplayground/vicuna_7b_v0',
|
16 |
+
'delta_ckpt_path': 'pretrained_ckpt/pandagpt_ckpt/7b/pytorch_model.pt',
|
17 |
+
'stage': 2,
|
18 |
+
'max_tgt_len': 128,
|
19 |
+
'lora_r': 32,
|
20 |
+
'lora_alpha': 32,
|
21 |
+
'lora_dropout': 0.1,
|
22 |
+
}
|
23 |
+
model = OpenLLAMAPEFTModel(**args)
|
24 |
+
delta_ckpt = torch.load(args['delta_ckpt_path'], map_location=torch.device('cpu'))
|
25 |
+
model.load_state_dict(delta_ckpt, strict=False)
|
26 |
+
model = model.half().cuda().eval() if torch.cuda.is_available() else model.eval()
|
27 |
+
print(f'[!] init the model over ...')
|
28 |
+
|
29 |
+
|
30 |
+
"""Override Chatbot.postprocess"""
|
31 |
+
|
32 |
+
|
33 |
+
def postprocess(self, y):
|
34 |
+
if y is None:
|
35 |
+
return []
|
36 |
+
for i, (message, response) in enumerate(y):
|
37 |
+
y[i] = (
|
38 |
+
None if message is None else mdtex2html.convert((message)),
|
39 |
+
None if response is None else mdtex2html.convert(response),
|
40 |
+
)
|
41 |
+
return y
|
42 |
+
|
43 |
+
|
44 |
+
gr.Chatbot.postprocess = postprocess
|
45 |
+
|
46 |
+
|
47 |
+
def parse_text(text):
|
48 |
+
"""copy from https://github.com/GaiZhenbiao/ChuanhuChatGPT/"""
|
49 |
+
lines = text.split("\n")
|
50 |
+
lines = [line for line in lines if line != ""]
|
51 |
+
count = 0
|
52 |
+
for i, line in enumerate(lines):
|
53 |
+
if "```" in line:
|
54 |
+
count += 1
|
55 |
+
items = line.split('`')
|
56 |
+
if count % 2 == 1:
|
57 |
+
lines[i] = f'<pre><code class="language-{items[-1]}">'
|
58 |
+
else:
|
59 |
+
lines[i] = f'<br></code></pre>'
|
60 |
+
else:
|
61 |
+
if i > 0:
|
62 |
+
if count % 2 == 1:
|
63 |
+
line = line.replace("`", "\`")
|
64 |
+
line = line.replace("<", "<")
|
65 |
+
line = line.replace(">", ">")
|
66 |
+
line = line.replace(" ", " ")
|
67 |
+
line = line.replace("*", "*")
|
68 |
+
line = line.replace("_", "_")
|
69 |
+
line = line.replace("-", "-")
|
70 |
+
line = line.replace(".", ".")
|
71 |
+
line = line.replace("!", "!")
|
72 |
+
line = line.replace("(", "(")
|
73 |
+
line = line.replace(")", ")")
|
74 |
+
line = line.replace("$", "$")
|
75 |
+
lines[i] = "<br>"+line
|
76 |
+
text = "".join(lines)
|
77 |
+
return text
|
78 |
+
|
79 |
+
|
80 |
+
def predict(
|
81 |
+
input,
|
82 |
+
image_path,
|
83 |
+
audio_path,
|
84 |
+
video_path,
|
85 |
+
thermal_path,
|
86 |
+
chatbot,
|
87 |
+
max_length,
|
88 |
+
top_p,
|
89 |
+
temperature,
|
90 |
+
history,
|
91 |
+
modality_cache,
|
92 |
+
):
|
93 |
+
if image_path is None and audio_path is None and video_path is None and thermal_path is None:
|
94 |
+
return [(input, "There is no image/audio/video provided. Please upload the file to start a conversation.")]
|
95 |
+
else:
|
96 |
+
print(f'[!] image path: {image_path}\n[!] audio path: {audio_path}\n[!] video path: {video_path}\n[!] thermal pah: {thermal_path}')
|
97 |
+
# prepare the prompt
|
98 |
+
prompt_text = ''
|
99 |
+
for idx, (q, a) in enumerate(history):
|
100 |
+
if idx == 0:
|
101 |
+
prompt_text += f'{q}\n### Assistant: {a}\n###'
|
102 |
+
else:
|
103 |
+
prompt_text += f' Human: {q}\n### Assistant: {a}\n###'
|
104 |
+
if len(history) == 0:
|
105 |
+
prompt_text += f'{input}'
|
106 |
+
else:
|
107 |
+
prompt_text += f' Human: {input}'
|
108 |
+
|
109 |
+
response = model.generate({
|
110 |
+
'prompt': prompt_text,
|
111 |
+
'image_paths': [image_path] if image_path else [],
|
112 |
+
'audio_paths': [audio_path] if audio_path else [],
|
113 |
+
'video_paths': [video_path] if video_path else [],
|
114 |
+
'thermal_paths': [thermal_path] if thermal_path else [],
|
115 |
+
'top_p': top_p,
|
116 |
+
'temperature': temperature,
|
117 |
+
'max_tgt_len': max_length,
|
118 |
+
'modality_embeds': modality_cache
|
119 |
+
})
|
120 |
+
chatbot.append((parse_text(input), parse_text(response)))
|
121 |
+
history.append((input, response))
|
122 |
+
return chatbot, history, modality_cache
|
123 |
+
|
124 |
+
|
125 |
+
def reset_user_input():
|
126 |
+
return gr.update(value='')
|
127 |
+
|
128 |
+
|
129 |
+
def reset_state():
|
130 |
+
return None, None, None, None, [], [], []
|
131 |
+
|
132 |
+
|
133 |
+
with gr.Blocks() as demo:
|
134 |
+
gr.HTML("""<h1 align="center">PandaGPT</h1>""")
|
135 |
+
gr.Markdown('''We note that the current online demo uses the 7B version of PandaGPT due to the limitation of computation resource.
|
136 |
+
|
137 |
+
Better results should be expected when switching to the 13B version of PandaGPT.
|
138 |
+
|
139 |
+
For more details on how to run 13B PandaGPT, please refer to our [main project repository](https://github.com/yxuansu/PandaGPT).''')
|
140 |
+
|
141 |
+
with gr.Row(scale=4):
|
142 |
+
with gr.Column(scale=2):
|
143 |
+
image_path = gr.Image(type="filepath", label="Image", value=None)
|
144 |
+
|
145 |
+
gr.Examples(
|
146 |
+
[
|
147 |
+
os.path.join(os.path.dirname(__file__), "assets/images/bird_image.jpg"),
|
148 |
+
os.path.join(os.path.dirname(__file__), "assets/images/dog_image.jpg"),
|
149 |
+
os.path.join(os.path.dirname(__file__), "assets/images/car_image.jpg"),
|
150 |
+
],
|
151 |
+
image_path
|
152 |
+
)
|
153 |
+
with gr.Column(scale=2):
|
154 |
+
audio_path = gr.Audio(type="filepath", label="Audio", value=None)
|
155 |
+
gr.Examples(
|
156 |
+
[
|
157 |
+
os.path.join(os.path.dirname(__file__), "assets/audios/bird_audio.wav"),
|
158 |
+
os.path.join(os.path.dirname(__file__), "assets/audios/dog_audio.wav"),
|
159 |
+
os.path.join(os.path.dirname(__file__), "assets/audios/car_audio.wav"),
|
160 |
+
],
|
161 |
+
audio_path
|
162 |
+
)
|
163 |
+
with gr.Row(scale=4):
|
164 |
+
with gr.Column(scale=2):
|
165 |
+
video_path = gr.Video(type='file', label="Video")
|
166 |
+
|
167 |
+
gr.Examples(
|
168 |
+
[
|
169 |
+
os.path.join(os.path.dirname(__file__), "assets/videos/world.mp4"),
|
170 |
+
os.path.join(os.path.dirname(__file__), "assets/videos/a.mp4"),
|
171 |
+
],
|
172 |
+
video_path
|
173 |
+
)
|
174 |
+
with gr.Column(scale=2):
|
175 |
+
thermal_path = gr.Image(type="filepath", label="Thermal Image", value=None)
|
176 |
+
|
177 |
+
gr.Examples(
|
178 |
+
[
|
179 |
+
os.path.join(os.path.dirname(__file__), "assets/thermals/190662.jpg"),
|
180 |
+
os.path.join(os.path.dirname(__file__), "assets/thermals/210009.jpg"),
|
181 |
+
],
|
182 |
+
thermal_path
|
183 |
+
)
|
184 |
+
|
185 |
+
chatbot = gr.Chatbot()
|
186 |
+
with gr.Row():
|
187 |
+
with gr.Column(scale=4):
|
188 |
+
with gr.Column(scale=12):
|
189 |
+
user_input = gr.Textbox(show_label=False, placeholder="Input...", lines=10).style(container=False)
|
190 |
+
with gr.Column(min_width=32, scale=1):
|
191 |
+
submitBtn = gr.Button("Submit", variant="primary")
|
192 |
+
with gr.Column(scale=1):
|
193 |
+
emptyBtn = gr.Button("Clear History")
|
194 |
+
max_length = gr.Slider(0, 512, value=128, step=1.0, label="Maximum length", interactive=True)
|
195 |
+
top_p = gr.Slider(0, 1, value=0.01, step=0.01, label="Top P", interactive=True)
|
196 |
+
temperature = gr.Slider(0, 1, value=0.8, step=0.01, label="Temperature", interactive=True)
|
197 |
+
|
198 |
+
history = gr.State([])
|
199 |
+
modality_cache = gr.State([])
|
200 |
+
|
201 |
+
submitBtn.click(
|
202 |
+
predict, [
|
203 |
+
user_input,
|
204 |
+
image_path,
|
205 |
+
audio_path,
|
206 |
+
video_path,
|
207 |
+
thermal_path,
|
208 |
+
chatbot,
|
209 |
+
max_length,
|
210 |
+
top_p,
|
211 |
+
temperature,
|
212 |
+
history,
|
213 |
+
modality_cache,
|
214 |
+
], [
|
215 |
+
chatbot,
|
216 |
+
history,
|
217 |
+
modality_cache
|
218 |
+
],
|
219 |
+
show_progress=True
|
220 |
+
)
|
221 |
+
|
222 |
+
submitBtn.click(reset_user_input, [], [user_input])
|
223 |
+
emptyBtn.click(reset_state, outputs=[
|
224 |
+
image_path,
|
225 |
+
audio_path,
|
226 |
+
video_path,
|
227 |
+
thermal_path,
|
228 |
+
chatbot,
|
229 |
+
history,
|
230 |
+
modality_cache
|
231 |
+
], show_progress=True)
|
232 |
+
|
233 |
+
|
234 |
+
demo.launch(enable_queue=True)
|