Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,234 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
2 |
+
import torch
|
3 |
+
|
4 |
+
# Load the model and tokenizer
|
5 |
+
model_name = "01-ai/Yi-34B-200K"
|
6 |
+
model = AutoModelForCausalLM.from_pretrained(model_name, trust_remote_code=True)
|
7 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
8 |
+
|
9 |
+
def run(message, chat_history, system_prompt, max_new_tokens=1024, temperature=0.3, top_p=0.9, top_k=50):
|
10 |
+
prompt = get_prompt(message, chat_history, system_prompt)
|
11 |
+
|
12 |
+
# Encode the prompt to tensor
|
13 |
+
input_ids = tokenizer.encode(prompt, return_tensors='pt')
|
14 |
+
|
15 |
+
# Generate a response using the model with adjusted parameters
|
16 |
+
response_ids = model.generate(
|
17 |
+
input_ids,
|
18 |
+
max_length=max_new_tokens + input_ids.shape[1],
|
19 |
+
temperature=temperature, # Controls randomness. Lower values make text more deterministic.
|
20 |
+
top_p=top_p, # Nucleus sampling: higher values allow more diversity.
|
21 |
+
top_k=top_k, # Top-k sampling: limits the number of top tokens considered.
|
22 |
+
pad_token_id=tokenizer.eos_token_id
|
23 |
+
)
|
24 |
+
|
25 |
+
# Decode the response
|
26 |
+
response = tokenizer.decode(response_ids[:, input_ids.shape[-1]:][0], skip_special_tokens=True)
|
27 |
+
return response
|
28 |
+
|
29 |
+
def get_prompt(message, chat_history, system_prompt):
|
30 |
+
texts = [f"<s>[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n"]
|
31 |
+
|
32 |
+
do_strip = False
|
33 |
+
for user_input, response in chat_history:
|
34 |
+
user_input = user_input.strip() if do_strip else user_input
|
35 |
+
do_strip = True
|
36 |
+
texts.append(f"{user_input} [/INST] {response.strip()} </s><s>[INST] ")
|
37 |
+
message = message.strip() if do_strip else message
|
38 |
+
texts.append(f"{message} [/INST]")
|
39 |
+
return ''.join(texts)
|
40 |
+
|
41 |
+
DEFAULT_SYSTEM_PROMPT = """
|
42 |
+
You are Yi. You are an AI assistant, you are moderately-polite and give only true information.
|
43 |
+
You carefully provide accurate, factual, thoughtful, nuanced answers, and are brilliant at reasoning.
|
44 |
+
If you think there might not be a correct answer, you say so. Since you are autoregressive,
|
45 |
+
each token you produce is another opportunity to use computation, therefore you always spend a few sentences explaining background context,
|
46 |
+
assumptions, and step-by-step thinking BEFORE you try to answer a question.
|
47 |
+
"""
|
48 |
+
MAX_MAX_NEW_TOKENS = 200000
|
49 |
+
DEFAULT_MAX_NEW_TOKENS = 100000
|
50 |
+
MAX_INPUT_TOKEN_LENGTH = 100000
|
51 |
+
|
52 |
+
DESCRIPTION = "# [Yi-6B](https://huggingface.co/01-ai/Yi-6B)"
|
53 |
+
|
54 |
+
def clear_and_save_textbox(message): return '', message
|
55 |
+
|
56 |
+
def display_input(message, history=[]):
|
57 |
+
history.append((message, ''))
|
58 |
+
return history
|
59 |
+
|
60 |
+
def delete_prev_fn(history=[]):
|
61 |
+
try:
|
62 |
+
message, _ = history.pop()
|
63 |
+
except IndexError:
|
64 |
+
message = ''
|
65 |
+
return history, message or ''
|
66 |
+
|
67 |
+
def generate(message, history_with_input, system_prompt, max_new_tokens, temperature, top_p, top_k):
|
68 |
+
if max_new_tokens > MAX_MAX_NEW_TOKENS:
|
69 |
+
raise ValueError
|
70 |
+
|
71 |
+
history = history_with_input[:-1]
|
72 |
+
generator = run(message, history, system_prompt, max_new_tokens, temperature, top_p, top_k)
|
73 |
+
try:
|
74 |
+
first_response = next(generator)
|
75 |
+
yield history + [(message, first_response)]
|
76 |
+
except StopIteration:
|
77 |
+
yield history + [(message, '')]
|
78 |
+
for response in generator:
|
79 |
+
yield history + [(message, response)]
|
80 |
+
|
81 |
+
def process_example(message):
|
82 |
+
generator = generate(message, [], DEFAULT_SYSTEM_PROMPT, 1024, 1, 0.95, 50)
|
83 |
+
for x in generator:
|
84 |
+
pass
|
85 |
+
return '', x
|
86 |
+
|
87 |
+
def check_input_token_length(message, chat_history, system_prompt):
|
88 |
+
input_token_length = len(message) + len(chat_history)
|
89 |
+
if input_token_length > MAX_INPUT_TOKEN_LENGTH:
|
90 |
+
raise gr.Error(f"The accumulated input is too long ({input_token_length} > {MAX_INPUT_TOKEN_LENGTH}). Clear your chat history and try again.")
|
91 |
+
|
92 |
+
with gr.Blocks(theme='ParityError/Anime') as demo:
|
93 |
+
gr.Markdown(DESCRIPTION)
|
94 |
+
|
95 |
+
|
96 |
+
|
97 |
+
with gr.Group():
|
98 |
+
chatbot = gr.Chatbot(label='Yi-6B')
|
99 |
+
with gr.Row():
|
100 |
+
textbox = gr.Textbox(
|
101 |
+
container=False,
|
102 |
+
show_label=False,
|
103 |
+
placeholder='Hi, Yi',
|
104 |
+
scale=10
|
105 |
+
)
|
106 |
+
submit_button = gr.Button('Submit', variant='primary', scale=1, min_width=0)
|
107 |
+
|
108 |
+
with gr.Row():
|
109 |
+
retry_button = gr.Button('Retry', variant='secondary')
|
110 |
+
undo_button = gr.Button('Undo', variant='secondary')
|
111 |
+
clear_button = gr.Button('Clear', variant='secondary')
|
112 |
+
|
113 |
+
saved_input = gr.State()
|
114 |
+
|
115 |
+
with gr.Accordion(label='Advanced options', open=False):
|
116 |
+
system_prompt = gr.Textbox(label='System prompt', value=DEFAULT_SYSTEM_PROMPT, lines=5, interactive=False)
|
117 |
+
max_new_tokens = gr.Slider(label='Max New Tokens', minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS)
|
118 |
+
temperature = gr.Slider(label='Temperature', minimum=0.1, maximum=4.0, step=0.1, value=0.1)
|
119 |
+
top_p = gr.Slider(label='Top-P (nucleus sampling)', minimum=0.05, maximum=1.0, step=0.05, value=0.9)
|
120 |
+
top_k = gr.Slider(label='Top-K', minimum=1, maximum=1000, step=1, value=10)
|
121 |
+
|
122 |
+
textbox.submit(
|
123 |
+
fn=clear_and_save_textbox,
|
124 |
+
inputs=textbox,
|
125 |
+
outputs=[textbox, saved_input],
|
126 |
+
api_name=False,
|
127 |
+
queue=False,
|
128 |
+
).then(
|
129 |
+
fn=display_input,
|
130 |
+
inputs=[saved_input, chatbot],
|
131 |
+
outputs=chatbot,
|
132 |
+
api_name=False,
|
133 |
+
queue=False,
|
134 |
+
).then(
|
135 |
+
fn=check_input_token_length,
|
136 |
+
inputs=[saved_input, chatbot, system_prompt],
|
137 |
+
api_name=False,
|
138 |
+
queue=False,
|
139 |
+
).success(
|
140 |
+
fn=generate,
|
141 |
+
inputs=[
|
142 |
+
saved_input,
|
143 |
+
chatbot,
|
144 |
+
system_prompt,
|
145 |
+
max_new_tokens,
|
146 |
+
temperature,
|
147 |
+
top_p,
|
148 |
+
top_k,
|
149 |
+
],
|
150 |
+
outputs=chatbot,
|
151 |
+
api_name=False,
|
152 |
+
)
|
153 |
+
|
154 |
+
button_event_preprocess = submit_button.click(
|
155 |
+
fn=clear_and_save_textbox,
|
156 |
+
inputs=textbox,
|
157 |
+
outputs=[textbox, saved_input],
|
158 |
+
api_name=False,
|
159 |
+
queue=False,
|
160 |
+
).then(
|
161 |
+
fn=display_input,
|
162 |
+
inputs=[saved_input, chatbot],
|
163 |
+
outputs=chatbot,
|
164 |
+
api_name=False,
|
165 |
+
queue=False,
|
166 |
+
).then(
|
167 |
+
fn=check_input_token_length,
|
168 |
+
inputs=[saved_input, chatbot, system_prompt],
|
169 |
+
api_name=False,
|
170 |
+
queue=False,
|
171 |
+
).success(
|
172 |
+
fn=generate,
|
173 |
+
inputs=[
|
174 |
+
saved_input,
|
175 |
+
chatbot,
|
176 |
+
system_prompt,
|
177 |
+
max_new_tokens,
|
178 |
+
temperature,
|
179 |
+
top_p,
|
180 |
+
top_k,
|
181 |
+
],
|
182 |
+
outputs=chatbot,
|
183 |
+
api_name=False,
|
184 |
+
)
|
185 |
+
|
186 |
+
retry_button.click(
|
187 |
+
fn=delete_prev_fn,
|
188 |
+
inputs=chatbot,
|
189 |
+
outputs=[chatbot, saved_input],
|
190 |
+
api_name=False,
|
191 |
+
queue=False,
|
192 |
+
).then(
|
193 |
+
fn=display_input,
|
194 |
+
inputs=[saved_input, chatbot],
|
195 |
+
outputs=chatbot,
|
196 |
+
api_name=False,
|
197 |
+
queue=False,
|
198 |
+
).then(
|
199 |
+
fn=generate,
|
200 |
+
inputs=[
|
201 |
+
saved_input,
|
202 |
+
chatbot,
|
203 |
+
system_prompt,
|
204 |
+
max_new_tokens,
|
205 |
+
temperature,
|
206 |
+
top_p,
|
207 |
+
top_k,
|
208 |
+
],
|
209 |
+
outputs=chatbot,
|
210 |
+
api_name=False,
|
211 |
+
)
|
212 |
+
|
213 |
+
undo_button.click(
|
214 |
+
fn=delete_prev_fn,
|
215 |
+
inputs=chatbot,
|
216 |
+
outputs=[chatbot, saved_input],
|
217 |
+
api_name=False,
|
218 |
+
queue=False,
|
219 |
+
).then(
|
220 |
+
fn=lambda x: x,
|
221 |
+
inputs=[saved_input],
|
222 |
+
outputs=textbox,
|
223 |
+
api_name=False,
|
224 |
+
queue=False,
|
225 |
+
)
|
226 |
+
|
227 |
+
clear_button.click(
|
228 |
+
fn=lambda: ([], ''),
|
229 |
+
outputs=[chatbot, saved_input],
|
230 |
+
queue=False,
|
231 |
+
api_name=False,
|
232 |
+
)
|
233 |
+
|
234 |
+
demo.queue(max_size=32).launch(show_api=False)
|