Spaces:
Sleeping
Sleeping
Update files to support multi-turn
Browse files
README.md
CHANGED
@@ -4,7 +4,7 @@ emoji: 💬
|
|
4 |
colorFrom: yellow
|
5 |
colorTo: purple
|
6 |
sdk: gradio
|
7 |
-
sdk_version: 5.0
|
8 |
app_file: app.py
|
9 |
pinned: true
|
10 |
license: apache-2.0
|
|
|
4 |
colorFrom: yellow
|
5 |
colorTo: purple
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 5.21.0
|
8 |
app_file: app.py
|
9 |
pinned: true
|
10 |
license: apache-2.0
|
app.py
CHANGED
@@ -11,14 +11,6 @@ from vllm import LLM, SamplingParams
|
|
11 |
import vllm
|
12 |
import re
|
13 |
|
14 |
-
from huggingface_hub import login
|
15 |
-
TOKEN = os.environ.get("TOKEN", None)
|
16 |
-
login(token=TOKEN)
|
17 |
-
|
18 |
-
print("transformers version:", transformers.__version__)
|
19 |
-
print("vllm version:", vllm.__version__)
|
20 |
-
print("gradio version:", gr.__version__)
|
21 |
-
|
22 |
|
23 |
def load_model_processor(model_path):
|
24 |
processor = AutoProcessor.from_pretrained(model_path)
|
@@ -32,24 +24,15 @@ def load_model_processor(model_path):
|
|
32 |
model_path1 = "SeaLLMs/SeaLLMs-Audio-7B"
|
33 |
model1, processor1 = load_model_processor(model_path1)
|
34 |
|
35 |
-
def
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
|
42 |
-
|
43 |
-
|
44 |
-
{"type": "text", "text": text},
|
45 |
-
]},]
|
46 |
-
else:
|
47 |
-
conversation = [
|
48 |
-
{"role": "user", "content": [
|
49 |
-
{"type": "audio", "audio_url": audio_url},
|
50 |
-
{"type": "text", "text": text},
|
51 |
-
]},]
|
52 |
-
|
53 |
text = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
|
54 |
audios = []
|
55 |
for message in conversation:
|
@@ -76,45 +59,76 @@ def response_to_audio(audio_url, text, model=None, processor=None, temperature =
|
|
76 |
|
77 |
output = model.generate([input], sampling_params=sampling_params)[0]
|
78 |
response = output.outputs[0].text
|
|
|
|
|
79 |
return response
|
80 |
|
81 |
-
def
|
82 |
-
|
83 |
|
84 |
def contains_chinese(text):
|
85 |
# Regular expression for Chinese characters
|
86 |
chinese_char_pattern = re.compile(r'[\u4e00-\u9fff]')
|
87 |
return bool(chinese_char_pattern.search(text))
|
88 |
|
89 |
-
def
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
with gr.Blocks() as demo:
|
100 |
-
# gr.Markdown(f"Evaluate {model_path1}")
|
101 |
gr.HTML("""<p align="center"><img src="https://DAMO-NLP-SG.github.io/SeaLLMs-Audio/static/images/seallm-audio-logo.png" style="height: 80px"/><p>""")
|
102 |
-
# gr.Image("images/seal_logo.png", elem_id="seal_logo", show_label=False,height=80,show_fullscreen_button=False)
|
103 |
gr.HTML("""<h1 align="center" id="space-title">SeaLLMs-Audio-Demo</h1>""")
|
104 |
-
# gr.Markdown(
|
105 |
-
# """\
|
106 |
-
# <center><font size=4>This WebUI is based on SeaLLMs-Audio-7B, developed by Alibaba DAMO Academy.<br>
|
107 |
-
# You can interact with the chatbot in <b>English, Chinese, Indonesian, Thai, or Vietnamese</b>.<br>
|
108 |
-
# For the input, you can input <b>audio and/or text</center>.""")
|
109 |
-
|
110 |
-
# # Links with proper formatting
|
111 |
-
# gr.Markdown(
|
112 |
-
# """<center><font size=4>
|
113 |
-
# <a href="https://huggingface.co/SeaLLMs/SeaLLMs-v3-7B-Chat">[Website]</a>
|
114 |
-
# <a href="https://huggingface.co/SeaLLMs/SeaLLMs-Audio-7B">[Model🤗]</a>
|
115 |
-
# <a href="https://github.com/DAMO-NLP-SG/SeaLLMs-Audio">[Github]</a>
|
116 |
-
# </center>""",
|
117 |
-
# )
|
118 |
|
119 |
gr.HTML(
|
120 |
"""<div style="text-align: center; font-size: 16px;">
|
@@ -141,42 +155,27 @@ with gr.Blocks() as demo:
|
|
141 |
# with gr.Column():
|
142 |
# repetition_penalty = gr.Slider(minimum=0, maximum=2, value=1.1, step=0.1, label="Repetition Penalty")
|
143 |
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
btn_submit = gr.Button("Submit")
|
154 |
-
btn_clear = gr.Button("Clear")
|
155 |
-
|
156 |
-
with gr.Row():
|
157 |
-
output_text1 = gr.Textbox(label=model_path1.split('/')[-1], interactive=False, elem_id="output_text1")
|
158 |
-
|
159 |
-
btn_submit.click(
|
160 |
-
fn=compare_responses,
|
161 |
-
inputs=[mic_input, additional_input],
|
162 |
-
outputs=[output_text1],
|
163 |
)
|
164 |
|
165 |
-
|
166 |
-
|
167 |
-
inputs=None,
|
168 |
-
outputs=[mic_input, additional_input, output_text1],
|
169 |
-
queue=False,
|
170 |
)
|
|
|
|
|
|
|
171 |
|
|
|
172 |
|
173 |
-
|
174 |
-
# share=False,
|
175 |
-
# inbrowser=True,
|
176 |
-
# server_port=7950,
|
177 |
-
# server_name="0.0.0.0",
|
178 |
-
# max_threads=40
|
179 |
-
# )
|
180 |
|
181 |
demo.launch(share=True)
|
182 |
demo.queue(default_concurrency_limit=40).launch(share=True)
|
|
|
11 |
import vllm
|
12 |
import re
|
13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
def load_model_processor(model_path):
|
16 |
processor = AutoProcessor.from_pretrained(model_path)
|
|
|
24 |
model_path1 = "SeaLLMs/SeaLLMs-Audio-7B"
|
25 |
model1, processor1 = load_model_processor(model_path1)
|
26 |
|
27 |
+
def response_to_audio_conv(conversation, model=None, processor=None, temperature = 0.7,repetition_penalty=1.1, top_p = 0.5,max_new_tokens = 2048):
|
28 |
+
turn = conversation[-1]
|
29 |
+
if turn["role"] == "user":
|
30 |
+
for content in turn['content']:
|
31 |
+
if content["type"] == "text":
|
32 |
+
if contains_chinese(content["text"]):
|
33 |
+
return "Caution! This demo does not support Chinese!"
|
34 |
+
|
35 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
36 |
text = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
|
37 |
audios = []
|
38 |
for message in conversation:
|
|
|
59 |
|
60 |
output = model.generate([input], sampling_params=sampling_params)[0]
|
61 |
response = output.outputs[0].text
|
62 |
+
if contains_chinese(response):
|
63 |
+
return "ERROR! Try a different instruction/prompt!"
|
64 |
return response
|
65 |
|
66 |
+
def print_like_dislike(x: gr.LikeData):
|
67 |
+
print(x.index, x.value, x.liked)
|
68 |
|
69 |
def contains_chinese(text):
|
70 |
# Regular expression for Chinese characters
|
71 |
chinese_char_pattern = re.compile(r'[\u4e00-\u9fff]')
|
72 |
return bool(chinese_char_pattern.search(text))
|
73 |
|
74 |
+
def add_message(history, message):
|
75 |
+
paths = []
|
76 |
+
for turn in history:
|
77 |
+
if turn['role'] == "user" and type(turn['content']) != str:
|
78 |
+
paths.append(turn['content'][0])
|
79 |
+
for x in message["files"]:
|
80 |
+
if x not in paths:
|
81 |
+
history.append({"role": "user", "content": {"path": x}})
|
82 |
+
if message["text"] is not None:
|
83 |
+
history.append({"role": "user", "content": message["text"]})
|
84 |
+
return history, gr.MultimodalTextbox(value=None, interactive=False)
|
85 |
+
|
86 |
+
def format_user_messgae(message):
|
87 |
+
if type(message['content']) == str:
|
88 |
+
return {"role": "user", "content": [{"type": "text", "text": message['content']}]}
|
89 |
+
else:
|
90 |
+
return {"role": "user", "content": [{"type": "audio", "audio_url": message['content'][0]}]}
|
91 |
+
|
92 |
+
def history_to_conversation(history):
|
93 |
+
conversation = []
|
94 |
+
audio_paths = []
|
95 |
+
for turn in history:
|
96 |
+
if turn['role'] == "user":
|
97 |
+
if not turn['content']:
|
98 |
+
continue
|
99 |
+
turn = format_user_messgae(turn)
|
100 |
+
if turn['content'][0]['type'] == 'audio':
|
101 |
+
if turn['content'][0]['audio_url'] in audio_paths:
|
102 |
+
continue
|
103 |
+
else:
|
104 |
+
audio_paths.append(turn['content'][0]['audio_url'])
|
105 |
+
|
106 |
+
if len(conversation) > 0 and conversation[-1]["role"] == "user":
|
107 |
+
conversation[-1]['content'].append(turn['content'][0])
|
108 |
+
else:
|
109 |
+
conversation.append(turn)
|
110 |
+
else:
|
111 |
+
conversation.append(turn)
|
112 |
|
113 |
+
print(json.dumps(conversation, indent=4, ensure_ascii=False))
|
114 |
+
return conversation
|
115 |
+
|
116 |
+
def bot(history: list, temperature = 0.7,repetition_penalty=1.1, top_p = 0.5,
|
117 |
+
max_new_tokens = 2048):
|
118 |
+
conversation = history_to_conversation(history)
|
119 |
+
response = response_to_audio_conv(conversation, model=model1, processor=processor1, temperature = temperature,repetition_penalty=repetition_penalty, top_p = top_p, max_new_tokens = max_new_tokens)
|
120 |
+
# response = "Nice to meet you!"
|
121 |
+
print("Bot:",response)
|
122 |
+
|
123 |
+
history.append({"role": "assistant", "content": ""})
|
124 |
+
for character in response:
|
125 |
+
history[-1]["content"] += character
|
126 |
+
time.sleep(0.01)
|
127 |
+
yield history
|
128 |
|
129 |
with gr.Blocks() as demo:
|
|
|
130 |
gr.HTML("""<p align="center"><img src="https://DAMO-NLP-SG.github.io/SeaLLMs-Audio/static/images/seallm-audio-logo.png" style="height: 80px"/><p>""")
|
|
|
131 |
gr.HTML("""<h1 align="center" id="space-title">SeaLLMs-Audio-Demo</h1>""")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
132 |
|
133 |
gr.HTML(
|
134 |
"""<div style="text-align: center; font-size: 16px;">
|
|
|
155 |
# with gr.Column():
|
156 |
# repetition_penalty = gr.Slider(minimum=0, maximum=2, value=1.1, step=0.1, label="Repetition Penalty")
|
157 |
|
158 |
+
chatbot = gr.Chatbot(elem_id="chatbot", bubble_full_width=False, type="messages")
|
159 |
+
|
160 |
+
chat_input = gr.MultimodalTextbox(
|
161 |
+
interactive=True,
|
162 |
+
file_count="single",
|
163 |
+
file_types=['.wav'],
|
164 |
+
placeholder="Enter message (optional) ...",
|
165 |
+
show_label=False,
|
166 |
+
sources=["microphone", "upload"],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
167 |
)
|
168 |
|
169 |
+
chat_msg = chat_input.submit(
|
170 |
+
add_message, [chatbot, chat_input], [chatbot, chat_input]
|
|
|
|
|
|
|
171 |
)
|
172 |
+
bot_msg = chat_msg.then(bot, chatbot, chatbot, api_name="bot_response")
|
173 |
+
# bot_msg = chat_msg.then(bot, [chatbot, temperature, repetition_penalty, top_p], chatbot, api_name="bot_response")
|
174 |
+
bot_msg.then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
|
175 |
|
176 |
+
# chatbot.like(print_like_dislike, None, None, like_user_message=True)
|
177 |
|
178 |
+
clear_button = gr.ClearButton([chatbot, chat_input])
|
|
|
|
|
|
|
|
|
|
|
|
|
179 |
|
180 |
demo.launch(share=True)
|
181 |
demo.queue(default_concurrency_limit=40).launch(share=True)
|