Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,206 @@
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
|
|
|
1 |
+
import os
|
2 |
+
|
3 |
import gradio as gr
|
4 |
+
from transformers import AutoTokenizer
|
5 |
+
|
6 |
+
|
7 |
+
DESCRIPTION = """
|
8 |
+
|
9 |
+
"""
|
10 |
+
|
11 |
+
LICENSE = """
|
12 |
+
|
13 |
+
"""
|
14 |
+
|
15 |
+
DEFAULT_SYSTEM_PROMPT = ""
|
16 |
+
|
17 |
+
API_URL = os.environ.get("API_URL")
|
18 |
+
TOKEN = os.environ.get("TOKEN")
|
19 |
+
|
20 |
+
HEADER = {
|
21 |
+
"accept": "application/json",
|
22 |
+
"Authorization": f"Bearer {TOKEN}", "Content-Type": "application/json",
|
23 |
+
}
|
24 |
+
|
25 |
+
MODEL_NAME="breeze-7b-instruct-v01"
|
26 |
+
TEMPERATURE=1
|
27 |
+
MAX_TOKENS=16
|
28 |
+
TOP_P=0
|
29 |
+
PRESENCE_PENALTY=0
|
30 |
+
FREQUENCY_PENALTY=0
|
31 |
+
|
32 |
+
|
33 |
+
eos_token = "</s>"
|
34 |
+
MAX_MAX_NEW_TOKENS = 4096
|
35 |
+
DEFAULT_MAX_NEW_TOKENS = 1536
|
36 |
+
|
37 |
+
max_prompt_length = 8192 - MAX_MAX_NEW_TOKENS - 10
|
38 |
+
|
39 |
+
model_name = "MediaTek-Research/Breeze-7B-Instruct-v0.1"
|
40 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
41 |
+
|
42 |
+
with gr.Blocks() as demo:
|
43 |
+
gr.Markdown(DESCRIPTION)
|
44 |
+
|
45 |
+
chatbot = gr.Chatbot()
|
46 |
+
with gr.Row():
|
47 |
+
msg = gr.Textbox(
|
48 |
+
container=False,
|
49 |
+
show_label=False,
|
50 |
+
placeholder='Type a message...',
|
51 |
+
scale=10,
|
52 |
+
)
|
53 |
+
submit_button = gr.Button('Submit',
|
54 |
+
variant='primary',
|
55 |
+
scale=1,
|
56 |
+
min_width=0)
|
57 |
+
|
58 |
+
with gr.Row():
|
59 |
+
retry_button = gr.Button('🔄 Retry', variant='secondary')
|
60 |
+
undo_button = gr.Button('↩️ Undo', variant='secondary')
|
61 |
+
clear = gr.Button('🗑️ Clear', variant='secondary')
|
62 |
+
|
63 |
+
saved_input = gr.State()
|
64 |
+
|
65 |
+
with gr.Accordion(label='Advanced options', open=False):
|
66 |
+
system_prompt = gr.Textbox(label='System prompt',
|
67 |
+
value=DEFAULT_SYSTEM_PROMPT,
|
68 |
+
lines=6)
|
69 |
+
max_new_tokens = gr.Slider(
|
70 |
+
label='Max new tokens',
|
71 |
+
minimum=1,
|
72 |
+
maximum=MAX_MAX_NEW_TOKENS,
|
73 |
+
step=1,
|
74 |
+
value=DEFAULT_MAX_NEW_TOKENS,
|
75 |
+
)
|
76 |
+
temperature = gr.Slider(
|
77 |
+
label='Temperature',
|
78 |
+
minimum=0.1,
|
79 |
+
maximum=1.0,
|
80 |
+
step=0.1,
|
81 |
+
value=0.3,
|
82 |
+
)
|
83 |
+
top_p = gr.Slider(
|
84 |
+
label='Top-p (nucleus sampling)',
|
85 |
+
minimum=0.05,
|
86 |
+
maximum=1.0,
|
87 |
+
step=0.05,
|
88 |
+
value=0.95,
|
89 |
+
)
|
90 |
+
top_k = gr.Slider(
|
91 |
+
label='Top-k',
|
92 |
+
minimum=1,
|
93 |
+
maximum=1000,
|
94 |
+
step=1,
|
95 |
+
value=50,
|
96 |
+
)
|
97 |
+
|
98 |
+
def user(user_message, history):
|
99 |
+
return "", history + [[user_message, None]]
|
100 |
+
|
101 |
+
|
102 |
+
def bot(history, max_new_tokens, temperature, top_p, top_k, system_prompt):
|
103 |
+
data = {
|
104 |
+
"model": MODEL_NAME,
|
105 |
+
"messages": str(message),
|
106 |
+
"temperature": TEMPERATURE,
|
107 |
+
"n": 1,
|
108 |
+
"max_tokens": MAX_TOKENS,
|
109 |
+
"stop": "",
|
110 |
+
"top_p": TOP_P,
|
111 |
+
"logprobs": 0,
|
112 |
+
"echo": False,
|
113 |
+
"presence_penalty": PRESENCE_PENALTY,
|
114 |
+
"frequency_penalty": FREQUENCY_PENALTY,
|
115 |
+
}
|
116 |
+
|
117 |
+
outputs = requests.post(url, headers=headers, data=json.dumps(data)).json()
|
118 |
+
return outputs
|
119 |
+
|
120 |
+
msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
|
121 |
+
fn=bot,
|
122 |
+
inputs=[
|
123 |
+
chatbot,
|
124 |
+
max_new_tokens,
|
125 |
+
temperature,
|
126 |
+
top_p,
|
127 |
+
top_k,
|
128 |
+
system_prompt,
|
129 |
+
],
|
130 |
+
outputs=chatbot
|
131 |
+
)
|
132 |
+
submit_button.click(
|
133 |
+
user, [msg, chatbot], [msg, chatbot], queue=False
|
134 |
+
).then(
|
135 |
+
fn=bot,
|
136 |
+
inputs=[
|
137 |
+
chatbot,
|
138 |
+
max_new_tokens,
|
139 |
+
temperature,
|
140 |
+
top_p,
|
141 |
+
top_k,
|
142 |
+
system_prompt,
|
143 |
+
],
|
144 |
+
outputs=chatbot
|
145 |
+
)
|
146 |
+
|
147 |
+
|
148 |
+
def delete_prev_fn(
|
149 |
+
history: list[tuple[str, str]]) -> tuple[list[tuple[str, str]], str]:
|
150 |
+
try:
|
151 |
+
message, _ = history.pop()
|
152 |
+
except IndexError:
|
153 |
+
message = ''
|
154 |
+
return history, message or ''
|
155 |
+
|
156 |
+
|
157 |
+
def display_input(message: str,
|
158 |
+
history: list[tuple[str, str]]) -> list[tuple[str, str]]:
|
159 |
+
history.append((message, ''))
|
160 |
+
return history
|
161 |
+
|
162 |
+
retry_button.click(
|
163 |
+
fn=delete_prev_fn,
|
164 |
+
inputs=chatbot,
|
165 |
+
outputs=[chatbot, saved_input],
|
166 |
+
api_name=False,
|
167 |
+
queue=False,
|
168 |
+
).then(
|
169 |
+
fn=display_input,
|
170 |
+
inputs=[saved_input, chatbot],
|
171 |
+
outputs=chatbot,
|
172 |
+
api_name=False,
|
173 |
+
queue=False,
|
174 |
+
).then(
|
175 |
+
fn=bot,
|
176 |
+
inputs=[
|
177 |
+
chatbot,
|
178 |
+
max_new_tokens,
|
179 |
+
temperature,
|
180 |
+
top_p,
|
181 |
+
top_k,
|
182 |
+
system_prompt,
|
183 |
+
],
|
184 |
+
outputs=chatbot,
|
185 |
+
)
|
186 |
+
|
187 |
+
undo_button.click(
|
188 |
+
fn=delete_prev_fn,
|
189 |
+
inputs=chatbot,
|
190 |
+
outputs=[chatbot, saved_input],
|
191 |
+
api_name=False,
|
192 |
+
queue=False,
|
193 |
+
).then(
|
194 |
+
fn=lambda x: x,
|
195 |
+
inputs=[saved_input],
|
196 |
+
outputs=msg,
|
197 |
+
api_name=False,
|
198 |
+
queue=False,
|
199 |
+
)
|
200 |
+
|
201 |
+
clear.click(lambda: None, None, chatbot, queue=False)
|
202 |
+
|
203 |
+
gr.Markdown(LICENSE)
|
204 |
|
205 |
+
demo.queue(concurrency_count=4, max_size=128)
|
206 |
+
demo.launch()
|