Spaces:
Runtime error
Runtime error
Dagfinn1962
commited on
Commit
•
38880eb
1
Parent(s):
c005425
Update app.py
Browse files
app.py
CHANGED
@@ -1,10 +1,9 @@
|
|
1 |
-
import openai
|
2 |
import time
|
3 |
import logging
|
4 |
import gradio as gr
|
5 |
-
|
6 |
from src.llm_boilers import llm_boiler
|
7 |
-
|
8 |
|
9 |
logging.basicConfig(format="%(asctime)s - %(message)s", level=logging.INFO)
|
10 |
logging.warning("READY. App started...")
|
@@ -12,24 +11,29 @@ logging.warning("READY. App started...")
|
|
12 |
|
13 |
class Chat:
|
14 |
default_system_prompt = "A conversation between a user and an LLM-based AI assistant. The assistant gives helpful and honest answers."
|
15 |
-
system_format = "system\n{}
|
16 |
|
17 |
-
def __init__(
|
|
|
|
|
18 |
if system is not None:
|
19 |
self.set_system_prompt(system)
|
20 |
else:
|
21 |
self.reset_system_prompt()
|
22 |
-
self.user = user if user else "user\n{}
|
23 |
-
self.assistant =
|
|
|
|
|
24 |
self.response_prefix = self.assistant.split("{}")[0]
|
25 |
|
26 |
def set_system_prompt(self, system_prompt):
|
|
|
27 |
return system_prompt
|
28 |
|
29 |
def reset_system_prompt(self):
|
30 |
return self.set_system_prompt(self.default_system_prompt)
|
31 |
|
32 |
-
def history_as_formatted_str(self, system, history):
|
33 |
system = self.system_format.format(system)
|
34 |
text = system + "".join(
|
35 |
[
|
@@ -44,18 +48,27 @@ class Chat:
|
|
44 |
)
|
45 |
text += self.user.format(history[-1][0])
|
46 |
text += self.response_prefix
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
51 |
|
52 |
return text
|
53 |
|
54 |
def clear_history(self, history):
|
55 |
return []
|
56 |
|
57 |
-
def turn(self, user_input: str
|
58 |
-
self.user_turn(user_input
|
59 |
return self.bot_turn()
|
60 |
|
61 |
def user_turn(self, user_input: str, history):
|
@@ -65,6 +78,8 @@ class Chat:
|
|
65 |
def bot_turn(self, system, history, openai_key):
|
66 |
conversation = self.history_as_formatted_str(system, history)
|
67 |
assistant_response = call_inf_server(conversation, openai_key)
|
|
|
|
|
68 |
history[-1][1] = ""
|
69 |
for chunk in assistant_response:
|
70 |
try:
|
@@ -76,18 +91,19 @@ class Chat:
|
|
76 |
|
77 |
|
78 |
def call_inf_server(prompt, openai_key):
|
79 |
-
model_id = "gpt-3.5-turbo"
|
80 |
model = llm_boiler(model_id, openai_key)
|
81 |
logging.warning(f'Inf via "{model_id}"" for prompt "{prompt}"')
|
82 |
|
83 |
try:
|
84 |
-
#
|
85 |
response = model.run(prompt, temperature=1.0)
|
86 |
logging.warning(f"Result of text generation: {response}")
|
87 |
return response
|
88 |
|
89 |
except Exception as e:
|
90 |
-
#
|
|
|
91 |
print(e)
|
92 |
time.sleep(2)
|
93 |
response = model.run(prompt, temperature=1.0)
|
@@ -95,21 +111,24 @@ def call_inf_server(prompt, openai_key):
|
|
95 |
return response
|
96 |
|
97 |
|
98 |
-
|
99 |
-
|
100 |
-
# Get the OpenAI key from the environment variable
|
101 |
-
openai_key = os.getenv("OPENAI_API_KEY")
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
|
106 |
gr.Markdown(
|
107 |
-
"""
|
108 |
-
|
109 |
-
|
110 |
-
"""
|
111 |
)
|
112 |
conversation = Chat()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
113 |
chatbot = gr.Chatbot().style(height=400)
|
114 |
with gr.Row():
|
115 |
with gr.Column():
|
@@ -138,48 +157,42 @@ with gr.Blocks(theme='HaleyCH/HaleyCH_Theme') as demo:
|
|
138 |
reset = gr.Button("Reset System Prompt")
|
139 |
with gr.Row():
|
140 |
gr.Markdown(
|
141 |
-
"Disclaimer: The gpt-3.5-turbo model can produce factually incorrect output and should not be solely relied on to produce "
|
142 |
"factually accurate information. The gpt-3.5-turbo model was trained on various public datasets; while great efforts "
|
143 |
"have been taken to clean the pretraining data, it is possible that this model could generate lewd, "
|
144 |
"biased, or otherwise offensive outputs.",
|
145 |
elem_classes=["disclaimer"],
|
146 |
)
|
147 |
-
with gr.Row():
|
148 |
-
gr.Markdown(
|
149 |
-
"[Privacy policy](https://gist.github.com/samhavens/c29c68cdcd420a9aa0202d0839876dac)",
|
150 |
-
elem_classes=["disclaimer"],
|
151 |
-
)submit_event = msg.submit(
|
152 |
-
fn=conversation.user_turn,
|
153 |
-
inputs=[msg],
|
154 |
-
outputs=[msg, chatbot],
|
155 |
-
queue=False,
|
156 |
-
).then(
|
157 |
-
fn=conversation.bot_turn,
|
158 |
-
inputs=[system, chatbot, openai_key],
|
159 |
-
outputs=[chatbot],
|
160 |
-
queue=True, # Change `queue=True` to `keep_in_queue=True`
|
161 |
-
)
|
162 |
-
|
163 |
-
submit_click_event = submit.click(
|
164 |
-
fn=conversation.user_turn,
|
165 |
-
inputs=[msg],
|
166 |
-
outputs=[msg, chatbot],
|
167 |
-
queue=False,
|
168 |
-
).then(
|
169 |
-
fn=conversation.bot_turn,
|
170 |
-
inputs=[system, chatbot, openai_key],
|
171 |
-
outputs=[chatbot],
|
172 |
-
queue=True, # Change `queue=True` to `keep_in_queue=True`
|
173 |
-
)
|
174 |
-
|
175 |
-
stop.click(
|
176 |
-
fn=None,
|
177 |
-
inputs=None,
|
178 |
-
outputs=None,
|
179 |
-
cancels=[submit_event, submit_click_event],
|
180 |
-
queue=False,
|
181 |
-
)
|
182 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
183 |
clear.click(lambda: None, None, chatbot, queue=False).then(
|
184 |
fn=conversation.clear_history,
|
185 |
inputs=[chatbot],
|
@@ -200,4 +213,4 @@ stop.click(
|
|
200 |
)
|
201 |
|
202 |
|
203 |
-
demo.queue(max_size=36, concurrency_count=14).launch(debug=True)
|
|
|
|
|
1 |
import time
|
2 |
import logging
|
3 |
import gradio as gr
|
4 |
+
|
5 |
from src.llm_boilers import llm_boiler
|
6 |
+
|
7 |
|
8 |
logging.basicConfig(format="%(asctime)s - %(message)s", level=logging.INFO)
|
9 |
logging.warning("READY. App started...")
|
|
|
11 |
|
12 |
class Chat:
|
13 |
default_system_prompt = "A conversation between a user and an LLM-based AI assistant. The assistant gives helpful and honest answers."
|
14 |
+
system_format = "<|im_start|>system\n{}<|im_end|>\n"
|
15 |
|
16 |
+
def __init__(
|
17 |
+
self, system: str = None, user: str = None, assistant: str = None
|
18 |
+
) -> None:
|
19 |
if system is not None:
|
20 |
self.set_system_prompt(system)
|
21 |
else:
|
22 |
self.reset_system_prompt()
|
23 |
+
self.user = user if user else "<|im_start|>user\n{}<|im_end|>\n"
|
24 |
+
self.assistant = (
|
25 |
+
assistant if assistant else "<|im_start|>assistant\n{}<|im_end|>\n"
|
26 |
+
)
|
27 |
self.response_prefix = self.assistant.split("{}")[0]
|
28 |
|
29 |
def set_system_prompt(self, system_prompt):
|
30 |
+
# self.system = self.system_format.format(system_prompt)
|
31 |
return system_prompt
|
32 |
|
33 |
def reset_system_prompt(self):
|
34 |
return self.set_system_prompt(self.default_system_prompt)
|
35 |
|
36 |
+
def history_as_formatted_str(self, system, history) -> str:
|
37 |
system = self.system_format.format(system)
|
38 |
text = system + "".join(
|
39 |
[
|
|
|
48 |
)
|
49 |
text += self.user.format(history[-1][0])
|
50 |
text += self.response_prefix
|
51 |
+
# stopgap solution to too long sequences
|
52 |
+
if len(text) > 4500:
|
53 |
+
# delete from the middle between <|im_start|> and <|im_end|>
|
54 |
+
# find the middle ones, then expand out
|
55 |
+
start = text.find("<|im_start|>", 139)
|
56 |
+
end = text.find("<|im_end|>", 139)
|
57 |
+
while end < len(text) and len(text) > 4500:
|
58 |
+
end = text.find("<|im_end|>", end + 1)
|
59 |
+
text = text[:start] + text[end + 1 :]
|
60 |
+
if len(text) > 4500:
|
61 |
+
# the nice way didn't work, just truncate
|
62 |
+
# deleting the beginning
|
63 |
+
text = text[-4500:]
|
64 |
|
65 |
return text
|
66 |
|
67 |
def clear_history(self, history):
|
68 |
return []
|
69 |
|
70 |
+
def turn(self, user_input: str):
|
71 |
+
self.user_turn(user_input)
|
72 |
return self.bot_turn()
|
73 |
|
74 |
def user_turn(self, user_input: str, history):
|
|
|
78 |
def bot_turn(self, system, history, openai_key):
|
79 |
conversation = self.history_as_formatted_str(system, history)
|
80 |
assistant_response = call_inf_server(conversation, openai_key)
|
81 |
+
# history[-1][-1] = assistant_response
|
82 |
+
# return history
|
83 |
history[-1][1] = ""
|
84 |
for chunk in assistant_response:
|
85 |
try:
|
|
|
91 |
|
92 |
|
93 |
def call_inf_server(prompt, openai_key):
|
94 |
+
model_id = "eolionross/gpt-3.5-turbo-demo" # "gpt-3.5-turbo-16k",
|
95 |
model = llm_boiler(model_id, openai_key)
|
96 |
logging.warning(f'Inf via "{model_id}"" for prompt "{prompt}"')
|
97 |
|
98 |
try:
|
99 |
+
# run text generation
|
100 |
response = model.run(prompt, temperature=1.0)
|
101 |
logging.warning(f"Result of text generation: {response}")
|
102 |
return response
|
103 |
|
104 |
except Exception as e:
|
105 |
+
# assume it is our error
|
106 |
+
# just wait and try one more time
|
107 |
print(e)
|
108 |
time.sleep(2)
|
109 |
response = model.run(prompt, temperature=1.0)
|
|
|
111 |
return response
|
112 |
|
113 |
|
114 |
+
with gr.Blocks(theme='ParityError/Anime') as demo:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
115 |
gr.Markdown(
|
116 |
+
"""<h3><center>Chat with gpt-3.5-turbo</center></h3>
|
117 |
+
|
118 |
+
"""
|
|
|
119 |
)
|
120 |
conversation = Chat()
|
121 |
+
with gr.Row():
|
122 |
+
with gr.Column():
|
123 |
+
# to do: change to openaikey input for public release
|
124 |
+
openai_key = gr.Textbox(
|
125 |
+
label="Open Ai key",
|
126 |
+
value="",
|
127 |
+
type="password",
|
128 |
+
placeholder="Enter your Open ai API KEY here ",
|
129 |
+
info="",
|
130 |
+
|
131 |
+
)
|
132 |
chatbot = gr.Chatbot().style(height=400)
|
133 |
with gr.Row():
|
134 |
with gr.Column():
|
|
|
157 |
reset = gr.Button("Reset System Prompt")
|
158 |
with gr.Row():
|
159 |
gr.Markdown(
|
160 |
+
"Disclaimer: The gpt-3.5-turbo model can produce factually incorrect output, and should not be solely relied on to produce "
|
161 |
"factually accurate information. The gpt-3.5-turbo model was trained on various public datasets; while great efforts "
|
162 |
"have been taken to clean the pretraining data, it is possible that this model could generate lewd, "
|
163 |
"biased, or otherwise offensive outputs.",
|
164 |
elem_classes=["disclaimer"],
|
165 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
166 |
|
167 |
+
submit_event = msg.submit(
|
168 |
+
fn=conversation.user_turn,
|
169 |
+
inputs=[msg, chatbot],
|
170 |
+
outputs=[msg, chatbot],
|
171 |
+
queue=False,
|
172 |
+
).then(
|
173 |
+
fn=conversation.bot_turn,
|
174 |
+
inputs=[system, chatbot, openai_key],
|
175 |
+
outputs=[chatbot],
|
176 |
+
queue=True,
|
177 |
+
)
|
178 |
+
submit_click_event = submit.click(
|
179 |
+
fn=conversation.user_turn,
|
180 |
+
inputs=[msg, chatbot],
|
181 |
+
outputs=[msg, chatbot],
|
182 |
+
queue=False,
|
183 |
+
).then(
|
184 |
+
fn=conversation.bot_turn,
|
185 |
+
inputs=[system, chatbot, openai_key],
|
186 |
+
outputs=[chatbot],
|
187 |
+
queue=True,
|
188 |
+
)
|
189 |
+
stop.click(
|
190 |
+
fn=None,
|
191 |
+
inputs=None,
|
192 |
+
outputs=None,
|
193 |
+
cancels=[submit_event, submit_click_event],
|
194 |
+
queue=False,
|
195 |
+
)
|
196 |
clear.click(lambda: None, None, chatbot, queue=False).then(
|
197 |
fn=conversation.clear_history,
|
198 |
inputs=[chatbot],
|
|
|
213 |
)
|
214 |
|
215 |
|
216 |
+
demo.queue(max_size=36, concurrency_count=14).launch(debug=True)
|