Update app.py
Browse files
app.py
CHANGED
@@ -1,68 +1,105 @@
|
|
1 |
-
import
|
2 |
-
|
3 |
-
import
|
4 |
-
import
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
-
|
7 |
-
|
8 |
-
model_name = "sberbank-ai/rugpt3small_based_on_gpt2"
|
9 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
10 |
-
model = AutoModelForCausalLM.from_pretrained(model_name)
|
11 |
-
return tokenizer, model
|
12 |
|
13 |
-
|
14 |
-
|
15 |
-
with torch.no_grad():
|
16 |
-
outputs = model.generate(inputs, max_length=100, num_return_sequences=1,
|
17 |
-
temperature=0.9, top_k=50, top_p=0.95)
|
18 |
-
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
19 |
-
return add_mistakes(response)
|
20 |
|
21 |
-
def
|
22 |
-
|
23 |
-
for i in range(len(words)):
|
24 |
-
if random.random() < 0.2: # 20% шанс ошибки в слове
|
25 |
-
words[i] = misspell_word(words[i])
|
26 |
-
return ' '.join(words)
|
27 |
|
28 |
-
def
|
29 |
-
if len(
|
30 |
-
|
31 |
-
|
32 |
-
consonants = 'бвгджзйклмнпрстфхцчшщ'
|
33 |
-
|
34 |
-
if random.random() < 0.5:
|
35 |
-
# Заменяем случайную гласную
|
36 |
-
for i, char in enumerate(word):
|
37 |
-
if char.lower() in vowels:
|
38 |
-
replacement = random.choice(vowels)
|
39 |
-
return word[:i] + replacement + word[i+1:]
|
40 |
-
else:
|
41 |
-
# Заменяем случайную согласную
|
42 |
-
for i, char in enumerate(word):
|
43 |
-
if char.lower() in consonants:
|
44 |
-
replacement = random.choice(consonants)
|
45 |
-
return word[:i] + replacement + word[i+1:]
|
46 |
-
return word
|
47 |
|
48 |
-
|
|
|
|
|
|
|
|
|
|
|
49 |
|
50 |
-
tokenizer, model = load_model()
|
51 |
|
52 |
-
|
53 |
-
|
|
|
|
|
|
|
|
|
|
|
54 |
|
55 |
-
for message in st.session_state.messages:
|
56 |
-
with st.chat_message(message["role"]):
|
57 |
-
st.markdown(message["content"])
|
58 |
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
|
68 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
os.system('pip install dashscope')
|
3 |
+
import gradio as gr
|
4 |
+
from http import HTTPStatus
|
5 |
+
import dashscope
|
6 |
+
from dashscope import Generation
|
7 |
+
from dashscope.api_entities.dashscope_response import Role
|
8 |
+
from typing import List, Optional, Tuple, Dict
|
9 |
+
from urllib.error import HTTPError
|
10 |
+
default_system = 'You are a helpful assistant.'
|
11 |
|
12 |
+
YOUR_API_TOKEN = os.getenv('YOUR_API_TOKEN')
|
13 |
+
dashscope.api_key = YOUR_API_TOKEN
|
|
|
|
|
|
|
|
|
14 |
|
15 |
+
History = List[Tuple[str, str]]
|
16 |
+
Messages = List[Dict[str, str]]
|
|
|
|
|
|
|
|
|
|
|
17 |
|
18 |
+
def clear_session() -> History:
|
19 |
+
return '', []
|
|
|
|
|
|
|
|
|
20 |
|
21 |
+
def modify_system_session(system: str) -> str:
|
22 |
+
if system is None or len(system) == 0:
|
23 |
+
system = default_system
|
24 |
+
return system, system, []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
+
def history_to_messages(history: History, system: str) -> Messages:
|
27 |
+
messages = [{'role': Role.SYSTEM, 'content': system}]
|
28 |
+
for h in history:
|
29 |
+
messages.append({'role': Role.USER, 'content': h[0]})
|
30 |
+
messages.append({'role': Role.ASSISTANT, 'content': h[1]})
|
31 |
+
return messages
|
32 |
|
|
|
33 |
|
34 |
+
def messages_to_history(messages: Messages) -> Tuple[str, History]:
|
35 |
+
assert messages[0]['role'] == Role.SYSTEM
|
36 |
+
system = messages[0]['content']
|
37 |
+
history = []
|
38 |
+
for q, r in zip(messages[1::2], messages[2::2]):
|
39 |
+
history.append([q['content'], r['content']])
|
40 |
+
return system, history
|
41 |
|
|
|
|
|
|
|
42 |
|
43 |
+
def model_chat(query: Optional[str], history: Optional[History], system: str
|
44 |
+
) -> Tuple[str, str, History]:
|
45 |
+
if query is None:
|
46 |
+
query = ''
|
47 |
+
if history is None:
|
48 |
+
history = []
|
49 |
+
messages = history_to_messages(history, system)
|
50 |
+
messages.append({'role': Role.USER, 'content': query})
|
51 |
+
gen = Generation.call(
|
52 |
+
model='qwen2-72b-instruct',
|
53 |
+
messages=messages,
|
54 |
+
result_format='message',
|
55 |
+
stream=True
|
56 |
+
)
|
57 |
+
for response in gen:
|
58 |
+
if response.status_code == HTTPStatus.OK:
|
59 |
+
role = response.output.choices[0].message.role
|
60 |
+
response = response.output.choices[0].message.content
|
61 |
+
system, history = messages_to_history(messages + [{'role': role, 'content': response}])
|
62 |
+
yield '', history, system
|
63 |
+
else:
|
64 |
+
raise ValueError('Request id: %s, Status code: %s, error code: %s, error message: %s' % (
|
65 |
+
response.request_id, response.status_code,
|
66 |
+
response.code, response.message
|
67 |
+
))
|
68 |
+
|
69 |
+
|
70 |
+
with gr.Blocks() as demo:
|
71 |
+
gr.Markdown("""<center><font size=8>Qwen2-72B-instruct Chat👾</center>""")
|
72 |
+
|
73 |
+
with gr.Row():
|
74 |
+
with gr.Column(scale=3):
|
75 |
+
system_input = gr.Textbox(value=default_system, lines=1, label='System')
|
76 |
+
with gr.Column(scale=1):
|
77 |
+
modify_system = gr.Button("🛠️ Set system prompt and clear history", scale=2)
|
78 |
+
system_state = gr.Textbox(value=default_system, visible=False)
|
79 |
+
chatbot = gr.Chatbot(label='qwen2-72B-instruct')
|
80 |
+
textbox = gr.Textbox(lines=1, label='Input')
|
81 |
+
|
82 |
+
with gr.Row():
|
83 |
+
clear_history = gr.Button("🧹 Clear history")
|
84 |
+
sumbit = gr.Button("🚀 Send")
|
85 |
+
|
86 |
+
textbox.submit(model_chat,
|
87 |
+
inputs=[textbox, chatbot, system_state],
|
88 |
+
outputs=[textbox, chatbot, system_input],
|
89 |
+
concurrency_limit = 40)
|
90 |
|
91 |
+
sumbit.click(model_chat,
|
92 |
+
inputs=[textbox, chatbot, system_state],
|
93 |
+
outputs=[textbox, chatbot, system_input],
|
94 |
+
concurrency_limit = 40)
|
95 |
+
clear_history.click(fn=clear_session,
|
96 |
+
inputs=[],
|
97 |
+
outputs=[textbox, chatbot],
|
98 |
+
concurrency_limit = 40)
|
99 |
+
modify_system.click(fn=modify_system_session,
|
100 |
+
inputs=[system_input],
|
101 |
+
outputs=[system_state, system_input, chatbot],
|
102 |
+
concurrency_limit = 40)
|
103 |
+
|
104 |
+
demo.queue(api_open=False)
|
105 |
+
demo.launch(max_threads=40)
|