split different users' data
Browse files- .gitignore +1 -0
- app.py +53 -47
- examples/cook_720p.mp4 +3 -0
- examples/references.txt +7 -0
- examples/temple_of_heaven_720p.mp4 +3 -0
- model/Vicuna.py +16 -91
.gitignore
CHANGED
@@ -3,6 +3,7 @@ output_*/
|
|
3 |
icl_inference_output/
|
4 |
.vscode/
|
5 |
tmp/
|
|
|
6 |
|
7 |
# Byte-compiled / optimized / DLL files
|
8 |
__pycache__/
|
|
|
3 |
icl_inference_output/
|
4 |
.vscode/
|
5 |
tmp/
|
6 |
+
gradio_cached_examples/
|
7 |
|
8 |
# Byte-compiled / optimized / DLL files
|
9 |
__pycache__/
|
app.py
CHANGED
@@ -1,4 +1,3 @@
|
|
1 |
-
import argparse
|
2 |
import time
|
3 |
|
4 |
import gradio as gr
|
@@ -7,64 +6,73 @@ from config.config_utils import get_config
|
|
7 |
from model import Captioner, VicunaHandler
|
8 |
|
9 |
|
10 |
-
def
|
11 |
-
return
|
12 |
|
13 |
|
14 |
-
def
|
15 |
-
|
16 |
-
return file_paths
|
17 |
|
18 |
|
19 |
-
def
|
20 |
-
|
21 |
-
return video
|
22 |
|
23 |
|
24 |
-
def respond(input, chat_history):
|
25 |
-
bot_response = handler.gr_chat(input)
|
26 |
chat_history.append((input, bot_response))
|
27 |
time.sleep(0.1)
|
28 |
-
return "", chat_history
|
29 |
-
|
30 |
-
|
31 |
-
def clear_chat(chat_history):
|
32 |
-
handler.chatbot.clear_conv_()
|
33 |
-
|
34 |
-
return "", []
|
35 |
-
|
36 |
|
37 |
|
|
|
38 |
config = get_config('config/infer.yaml')
|
39 |
-
|
40 |
-
captioner = Captioner(config) # global
|
41 |
-
|
42 |
-
global handler
|
43 |
handler = VicunaHandler(config['vicuna'])
|
44 |
|
45 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
46 |
-
gr.Markdown(
|
47 |
-
|
48 |
-
|
49 |
-
""")
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
with gr.Row():
|
51 |
with gr.Column():
|
52 |
video_path = gr.Video(label="Video")
|
53 |
|
54 |
with gr.Column():
|
55 |
-
upload_button = gr.Button(
|
56 |
-
|
57 |
-
|
58 |
num_frames = gr.Slider(
|
59 |
minimum=5,
|
60 |
value=12,
|
61 |
maximum=12,
|
62 |
step=1,
|
63 |
-
label="Number of frames
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
64 |
|
65 |
with gr.Column():
|
|
|
66 |
chatbot = gr.Chatbot()
|
67 |
-
|
|
|
68 |
with gr.Row(visible=False) as input:
|
69 |
with gr.Column(scale=0.7):
|
70 |
txt = gr.Textbox(
|
@@ -76,22 +84,20 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
|
76 |
with gr.Column(scale=0.15, min_width=0):
|
77 |
clear_button = gr.Button("CLEAR")
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
txt.submit(respond, inputs=[txt, chatbot], outputs=[txt, chatbot])
|
92 |
run_button.click(
|
93 |
-
respond, inputs=[txt, chatbot], outputs=[txt, chatbot])
|
94 |
clear_button.click(
|
95 |
-
clear_chat, inputs=[
|
96 |
|
97 |
demo.launch()
|
|
|
|
|
1 |
import time
|
2 |
|
3 |
import gradio as gr
|
|
|
6 |
from model import Captioner, VicunaHandler
|
7 |
|
8 |
|
9 |
+
def mirror(x):
|
10 |
+
return x
|
11 |
|
12 |
|
13 |
+
def clear_chat(conv_template):
|
14 |
+
return "", [], conv_template
|
|
|
15 |
|
16 |
|
17 |
+
def clear_four():
|
18 |
+
return [], [], [], []
|
|
|
19 |
|
20 |
|
21 |
+
def respond(input, chat_history, conv):
|
22 |
+
bot_response, new_conv = handler.gr_chat(input, conv)
|
23 |
chat_history.append((input, bot_response))
|
24 |
time.sleep(0.1)
|
25 |
+
return "", chat_history, new_conv
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
27 |
|
28 |
+
# global variables
|
29 |
config = get_config('config/infer.yaml')
|
30 |
+
captioner = Captioner(config)
|
|
|
|
|
|
|
31 |
handler = VicunaHandler(config['vicuna'])
|
32 |
|
33 |
with gr.Blocks(theme=gr.themes.Soft()) as demo:
|
34 |
+
gr.Markdown(
|
35 |
+
"## <h1><center><img src='https://github.com/InvincibleWyq/ChatVID/assets/37479394/1a7f47ca-ffbd-4720-b43a-4304fcaa8657' height=40/> ChatVID</center></h1>"
|
36 |
+
)
|
37 |
+
gr.Markdown("""🔥 [ChatVID](https://github.com/InvincibleWyq/ChatVID) is a
|
38 |
+
video chatbot. Please give us a ⭐ Star!""")
|
39 |
+
gr.Markdown("""🎥 You may use the example video by clicking it.""")
|
40 |
+
gr.Markdown("""🚀 For any questions or suggestions, feel free to drop Yiqin
|
41 |
+
an email at <a href="mailto:[email protected]">[email protected]</a>
|
42 |
+
or open an issue.""")
|
43 |
+
|
44 |
with gr.Row():
|
45 |
with gr.Column():
|
46 |
video_path = gr.Video(label="Video")
|
47 |
|
48 |
with gr.Column():
|
49 |
+
upload_button = gr.Button("""Upload & Process.
|
50 |
+
(Click and wait 3min until dialog box appears)""")
|
51 |
+
|
52 |
num_frames = gr.Slider(
|
53 |
minimum=5,
|
54 |
value=12,
|
55 |
maximum=12,
|
56 |
step=1,
|
57 |
+
label="Number of frames")
|
58 |
+
|
59 |
+
gr.Markdown("## Video Examples")
|
60 |
+
gr.Examples(
|
61 |
+
examples=[
|
62 |
+
"examples/cook_720p.mp4",
|
63 |
+
"examples/temple_of_heaven_720p.mp4"
|
64 |
+
],
|
65 |
+
inputs=video_path,
|
66 |
+
outputs=video_path,
|
67 |
+
fn=mirror,
|
68 |
+
cache_examples=True,
|
69 |
+
)
|
70 |
|
71 |
with gr.Column():
|
72 |
+
caption_box = gr.Textbox("")
|
73 |
chatbot = gr.Chatbot()
|
74 |
+
conv_template = gr.State("") # determined by the video
|
75 |
+
conv = gr.State("") # updated thourghout the conversation
|
76 |
with gr.Row(visible=False) as input:
|
77 |
with gr.Column(scale=0.7):
|
78 |
txt = gr.Textbox(
|
|
|
84 |
with gr.Column(scale=0.15, min_width=0):
|
85 |
clear_button = gr.Button("CLEAR")
|
86 |
|
87 |
+
# conv_template and conv are `Conversation` objects
|
88 |
+
upload_button.click(lambda: gr.update(visible=False), None, input).then(
|
89 |
+
clear_four, None, [chatbot, conv, conv_template, caption_box]).then(
|
90 |
+
captioner.caption_video, [video_path, num_frames],
|
91 |
+
[conv_template]).then(mirror, [conv_template], [caption_box]).then(
|
92 |
+
handler.gr_chatbot_init, [conv_template],
|
93 |
+
[conv_template, conv]).then(lambda: gr.update(visible=True),
|
94 |
+
None, input)
|
95 |
+
|
96 |
+
txt.submit(
|
97 |
+
respond, inputs=[txt, chatbot, conv], outputs=[txt, chatbot, conv])
|
|
|
|
|
98 |
run_button.click(
|
99 |
+
respond, inputs=[txt, chatbot, conv], outputs=[txt, chatbot, conv])
|
100 |
clear_button.click(
|
101 |
+
clear_chat, inputs=[conv_template], outputs=[txt, chatbot, conv])
|
102 |
|
103 |
demo.launch()
|
examples/cook_720p.mp4
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:aa232686c22066b90fe099e9bb4f0ad093693685368eb7590ddd843deb40f574
|
3 |
+
size 5320367
|
examples/references.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
The video [Temple of Heaven - UNESCO World Heritage Site] by YouTube creator [World Heritage Journey] used under Fair Use.
|
2 |
+
Link:
|
3 |
+
https://www.youtube.com/watch?v=9xLoyYY_5rc
|
4 |
+
|
5 |
+
The video [做饭糊弄学 十分钟晚餐 今天吃 :番茄西兰花炒蛋] from Bilibili user [香蕉柿子梨] used under Fair Use.
|
6 |
+
Link:
|
7 |
+
https://www.bilibili.com/video/BV1RY411e74Z
|
examples/temple_of_heaven_720p.mp4
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:e60ce16122b0a6277c10efc5c37cc9b89c963913b2d382539fb7cc101bbd0851
|
3 |
+
size 4217766
|
model/Vicuna.py
CHANGED
@@ -1,8 +1,6 @@
|
|
1 |
from model.fastchat.conversation import (Conversation, SeparatorStyle,
|
2 |
-
compute_skip_echo_len
|
3 |
-
|
4 |
-
from model.fastchat.serve.inference import (ChatIO, chat_loop, generate_stream,
|
5 |
-
load_model)
|
6 |
|
7 |
|
8 |
class SimpleChatIO(ChatIO):
|
@@ -35,7 +33,6 @@ class VicunaChatBot:
|
|
35 |
num_gpus: str,
|
36 |
max_gpu_memory: str,
|
37 |
load_8bit: bool,
|
38 |
-
conv_template,
|
39 |
ChatIO: ChatIO,
|
40 |
debug: bool,
|
41 |
):
|
@@ -48,25 +45,18 @@ class VicunaChatBot:
|
|
48 |
num_gpus, max_gpu_memory,
|
49 |
load_8bit, debug)
|
50 |
|
51 |
-
|
52 |
-
|
53 |
-
else:
|
54 |
-
self.conv = get_default_conv_template(model_path).copy()
|
55 |
-
|
56 |
-
self.conv_template = self.conv.copy()
|
57 |
-
|
58 |
-
def chat(self, inp: str, temperature: float, max_new_tokens: int):
|
59 |
""" Vicuna as a chatbot. """
|
60 |
-
|
61 |
-
|
62 |
|
63 |
generate_stream_func = generate_stream
|
64 |
-
prompt =
|
65 |
|
66 |
-
skip_echo_len = compute_skip_echo_len(self.model_path,
|
67 |
-
prompt)
|
68 |
stop_str = (
|
69 |
-
|
70 |
in [SeparatorStyle.SINGLE, SeparatorStyle.BAIZE] else None)
|
71 |
params = {
|
72 |
"model": self.model_path,
|
@@ -76,65 +66,13 @@ class VicunaChatBot:
|
|
76 |
"stop": stop_str,
|
77 |
}
|
78 |
print(prompt)
|
79 |
-
self.chatio.prompt_for_output(
|
80 |
output_stream = generate_stream_func(self.model, self.tokenizer,
|
81 |
params, self.device)
|
82 |
outputs = self.chatio.stream_output(output_stream, skip_echo_len)
|
83 |
# NOTE: strip is important to align with the training data.
|
84 |
-
|
85 |
-
return outputs
|
86 |
-
|
87 |
-
def summarise(self, caption: dict, temperature: float,
|
88 |
-
max_new_tokens: int):
|
89 |
-
""" Vicuna as a summariser. """
|
90 |
-
questions = caption
|
91 |
-
captions = {}
|
92 |
-
for id, question in questions.items():
|
93 |
-
# Reset the conversation for each iteration
|
94 |
-
self.conv = get_default_conv_template(self.model_path).copy()
|
95 |
-
self.conv.append_message(self.conv.roles[0], question)
|
96 |
-
self.conv.append_message(self.conv.roles[1], None)
|
97 |
-
|
98 |
-
generate_stream_func = generate_stream
|
99 |
-
prompt = self.conv.get_prompt()
|
100 |
-
|
101 |
-
skip_echo_len = compute_skip_echo_len(self.model_path, self.conv,
|
102 |
-
prompt)
|
103 |
-
stop_str = (
|
104 |
-
self.conv.sep if self.conv.sep_style
|
105 |
-
in [SeparatorStyle.SINGLE, SeparatorStyle.BAIZE] else None)
|
106 |
-
|
107 |
-
params = {
|
108 |
-
"model": self.model_path,
|
109 |
-
"prompt": prompt,
|
110 |
-
"temperature": temperature,
|
111 |
-
"max_new_tokens": max_new_tokens,
|
112 |
-
"stop": stop_str,
|
113 |
-
}
|
114 |
-
|
115 |
-
self.chatio.prompt_for_output(self.conv.roles[1])
|
116 |
-
output_stream = generate_stream_func(self.model, self.tokenizer,
|
117 |
-
params, self.device)
|
118 |
-
outputs = self.chatio.stream_output(output_stream, skip_echo_len)
|
119 |
-
captions[id] = outputs
|
120 |
-
|
121 |
-
if self.debug:
|
122 |
-
print("\n", {"prompt": prompt, "outputs": outputs}, "\n")
|
123 |
-
|
124 |
-
print(captions)
|
125 |
-
return captions
|
126 |
-
|
127 |
-
def clear_conv_(self):
|
128 |
-
""" Clear the conversation. """
|
129 |
-
self.conv = self.conv_template.copy()
|
130 |
-
|
131 |
-
def change_conv_template_(self, conv_template):
|
132 |
-
self.conv_template = conv_template.copy()
|
133 |
-
self.conv = conv_template.copy()
|
134 |
-
|
135 |
-
def change_conv_(self, conv_template):
|
136 |
-
""" Change the conversation. """
|
137 |
-
self.conv = conv_template.copy()
|
138 |
|
139 |
|
140 |
class VicunaHandler:
|
@@ -150,38 +88,25 @@ class VicunaHandler:
|
|
150 |
self.config['num_gpus'],
|
151 |
self.config['max_gpu_memory'],
|
152 |
self.config['load_8bit'],
|
153 |
-
None,
|
154 |
self.chat_io,
|
155 |
self.config['debug'],
|
156 |
)
|
157 |
|
158 |
def chat(self):
|
159 |
""" Chat with the Vicuna. """
|
160 |
-
|
161 |
-
chat_loop(
|
162 |
-
self.config['model_path'],
|
163 |
-
self.config['device'],
|
164 |
-
self.config['num_gpus'],
|
165 |
-
self.config['max_gpu_memory'],
|
166 |
-
self.config['load_8bit'],
|
167 |
-
template,
|
168 |
-
self.config['temperature'],
|
169 |
-
self.config['max_new_tokens'],
|
170 |
-
self.chat_io,
|
171 |
-
self.config['debug'],
|
172 |
-
)
|
173 |
|
174 |
def gr_chatbot_init(self, caption: str):
|
175 |
""" Initialise the chatbot for gradio. """
|
176 |
|
177 |
template = self._construct_conversation(caption)
|
178 |
-
self.chatbot.change_conv_template_(template)
|
179 |
print("Chatbot initialised.")
|
|
|
180 |
|
181 |
-
def gr_chat(self, inp):
|
182 |
""" Chat using gradio as the frontend. """
|
183 |
return self.chatbot.chat(inp, self.config['temperature'],
|
184 |
-
self.config['max_new_tokens'])
|
185 |
|
186 |
def _construct_conversation(self, prompt):
|
187 |
""" Construct a conversation template.
|
|
|
1 |
from model.fastchat.conversation import (Conversation, SeparatorStyle,
|
2 |
+
compute_skip_echo_len)
|
3 |
+
from model.fastchat.serve.inference import ChatIO, generate_stream, load_model
|
|
|
|
|
4 |
|
5 |
|
6 |
class SimpleChatIO(ChatIO):
|
|
|
33 |
num_gpus: str,
|
34 |
max_gpu_memory: str,
|
35 |
load_8bit: bool,
|
|
|
36 |
ChatIO: ChatIO,
|
37 |
debug: bool,
|
38 |
):
|
|
|
45 |
num_gpus, max_gpu_memory,
|
46 |
load_8bit, debug)
|
47 |
|
48 |
+
def chat(self, inp: str, temperature: float, max_new_tokens: int,
|
49 |
+
conv: Conversation):
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
""" Vicuna as a chatbot. """
|
51 |
+
conv.append_message(conv.roles[0], inp)
|
52 |
+
conv.append_message(conv.roles[1], None)
|
53 |
|
54 |
generate_stream_func = generate_stream
|
55 |
+
prompt = conv.get_prompt()
|
56 |
|
57 |
+
skip_echo_len = compute_skip_echo_len(self.model_path, conv, prompt)
|
|
|
58 |
stop_str = (
|
59 |
+
conv.sep if conv.sep_style
|
60 |
in [SeparatorStyle.SINGLE, SeparatorStyle.BAIZE] else None)
|
61 |
params = {
|
62 |
"model": self.model_path,
|
|
|
66 |
"stop": stop_str,
|
67 |
}
|
68 |
print(prompt)
|
69 |
+
self.chatio.prompt_for_output(conv.roles[1])
|
70 |
output_stream = generate_stream_func(self.model, self.tokenizer,
|
71 |
params, self.device)
|
72 |
outputs = self.chatio.stream_output(output_stream, skip_echo_len)
|
73 |
# NOTE: strip is important to align with the training data.
|
74 |
+
conv.messages[-1][-1] = outputs.strip()
|
75 |
+
return outputs, conv
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
|
77 |
|
78 |
class VicunaHandler:
|
|
|
88 |
self.config['num_gpus'],
|
89 |
self.config['max_gpu_memory'],
|
90 |
self.config['load_8bit'],
|
|
|
91 |
self.chat_io,
|
92 |
self.config['debug'],
|
93 |
)
|
94 |
|
95 |
def chat(self):
|
96 |
""" Chat with the Vicuna. """
|
97 |
+
pass
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
def gr_chatbot_init(self, caption: str):
|
100 |
""" Initialise the chatbot for gradio. """
|
101 |
|
102 |
template = self._construct_conversation(caption)
|
|
|
103 |
print("Chatbot initialised.")
|
104 |
+
return template.copy(), template.copy()
|
105 |
|
106 |
+
def gr_chat(self, inp, conv: Conversation):
|
107 |
""" Chat using gradio as the frontend. """
|
108 |
return self.chatbot.chat(inp, self.config['temperature'],
|
109 |
+
self.config['max_new_tokens'], conv)
|
110 |
|
111 |
def _construct_conversation(self, prompt):
|
112 |
""" Construct a conversation template.
|