Spaces:
Running
Running
github-actions
commited on
Commit
·
a70549c
1
Parent(s):
16848ed
Sync updates from source repository
Browse files
app.py
CHANGED
|
@@ -1,18 +1,18 @@
|
|
| 1 |
from omegaconf import OmegaConf
|
| 2 |
from query import VectaraQuery
|
| 3 |
import os
|
| 4 |
-
import
|
| 5 |
-
import json
|
| 6 |
import uuid
|
| 7 |
|
| 8 |
import streamlit as st
|
| 9 |
from streamlit_pills import pills
|
| 10 |
from streamlit_feedback import streamlit_feedback
|
| 11 |
|
| 12 |
-
from
|
|
|
|
| 13 |
|
| 14 |
max_examples = 6
|
| 15 |
-
languages = {'English': 'eng', 'Spanish': 'spa', 'French': '
|
| 16 |
'Portuguese': 'por', 'Italian': 'ita', 'Japanese': 'jpn', 'Korean': 'kor', 'Russian': 'rus', 'Turkish': 'tur', 'Persian (Farsi)': 'fas',
|
| 17 |
'Vietnamese': 'vie', 'Thai': 'tha', 'Hebrew': 'heb', 'Dutch': 'nld', 'Indonesian': 'ind', 'Polish': 'pol', 'Ukrainian': 'ukr',
|
| 18 |
'Romanian': 'ron', 'Swedish': 'swe', 'Czech': 'ces', 'Greek': 'ell', 'Bengali': 'ben', 'Malay (or Malaysian)': 'msa', 'Urdu': 'urd'}
|
|
@@ -21,36 +21,6 @@ languages = {'English': 'eng', 'Spanish': 'spa', 'French': 'frs', 'Chinese': 'zh
|
|
| 21 |
if 'device_id' not in st.session_state:
|
| 22 |
st.session_state.device_id = str(uuid.uuid4())
|
| 23 |
|
| 24 |
-
headers = {
|
| 25 |
-
'Content-Type': 'application/json',
|
| 26 |
-
'Accept': '*/*'
|
| 27 |
-
}
|
| 28 |
-
amp_api_key = os.getenv('AMPLITUDE_TOKEN')
|
| 29 |
-
|
| 30 |
-
def thumbs_feedback(feedback, **kwargs):
|
| 31 |
-
"""
|
| 32 |
-
Sends feedback to Amplitude Analytics
|
| 33 |
-
"""
|
| 34 |
-
data = {
|
| 35 |
-
"api_key": amp_api_key,
|
| 36 |
-
"events": [{
|
| 37 |
-
"device_id": st.session_state.device_id,
|
| 38 |
-
"event_type": "provided_feedback",
|
| 39 |
-
"event_properties": {
|
| 40 |
-
"Space Name": kwargs.get("title", "Unknown Space Name"),
|
| 41 |
-
"Demo Type": "chatbot",
|
| 42 |
-
"query": kwargs.get("prompt", "No user input"),
|
| 43 |
-
"response": kwargs.get("response", "No chat response"),
|
| 44 |
-
"feedback": feedback["score"],
|
| 45 |
-
"Response Language": st.session_state.language
|
| 46 |
-
}
|
| 47 |
-
}]
|
| 48 |
-
}
|
| 49 |
-
response = requests.post('https://api2.amplitude.com/2/httpapi', headers=headers, data=json.dumps(data))
|
| 50 |
-
if response.status_code != 200:
|
| 51 |
-
print(f"Request failed with status code {response.status_code}. Response Text: {response.text}")
|
| 52 |
-
|
| 53 |
-
st.session_state.feedback_key += 1
|
| 54 |
|
| 55 |
if "feedback_key" not in st.session_state:
|
| 56 |
st.session_state.feedback_key = 0
|
|
@@ -61,6 +31,12 @@ def isTrue(x) -> bool:
|
|
| 61 |
return x.strip().lower() == 'true'
|
| 62 |
|
| 63 |
def launch_bot():
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 64 |
def generate_response(question):
|
| 65 |
response = vq.submit_query(question, languages[st.session_state.language])
|
| 66 |
return response
|
|
@@ -113,9 +89,16 @@ def launch_bot():
|
|
| 113 |
cfg.language = st.selectbox('Language:', languages.keys())
|
| 114 |
if st.session_state.language != cfg.language:
|
| 115 |
st.session_state.language = cfg.language
|
| 116 |
-
|
| 117 |
st.rerun()
|
| 118 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 119 |
st.markdown("---")
|
| 120 |
st.markdown(
|
| 121 |
"## How this works?\n"
|
|
@@ -129,7 +112,7 @@ def launch_bot():
|
|
| 129 |
st.markdown(f"<center> <h2> Vectara AI Assistant: {cfg.title} </h2> </center>", unsafe_allow_html=True)
|
| 130 |
|
| 131 |
if "messages" not in st.session_state.keys():
|
| 132 |
-
|
| 133 |
|
| 134 |
# Display chat messages
|
| 135 |
for message in st.session_state.messages:
|
|
@@ -157,40 +140,32 @@ def launch_bot():
|
|
| 157 |
if st.session_state.messages[-1]["role"] != "assistant":
|
| 158 |
with st.chat_message("assistant"):
|
| 159 |
if cfg.streaming:
|
| 160 |
-
stream = generate_streaming_response(prompt)
|
| 161 |
-
response = st.write_stream(stream)
|
| 162 |
else:
|
| 163 |
with st.spinner("Thinking..."):
|
| 164 |
response = generate_response(prompt)
|
| 165 |
st.write(response)
|
|
|
|
|
|
|
| 166 |
message = {"role": "assistant", "content": response}
|
| 167 |
st.session_state.messages.append(message)
|
| 168 |
|
| 169 |
# Send query and response to Amplitude Analytics
|
| 170 |
-
|
| 171 |
-
"
|
| 172 |
-
"
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
"Space Name": cfg["title"],
|
| 177 |
-
"Demo Type": "chatbot",
|
| 178 |
-
"query": st.session_state.messages[-2]["content"],
|
| 179 |
-
"response": st.session_state.messages[-1]["content"],
|
| 180 |
-
"Response Language": st.session_state.language
|
| 181 |
-
}
|
| 182 |
-
}]
|
| 183 |
-
}
|
| 184 |
-
response = requests.post('https://api2.amplitude.com/2/httpapi', headers=headers, data=json.dumps(data))
|
| 185 |
-
if response.status_code != 200:
|
| 186 |
-
print(f"Amplitude request failed with status code {response.status_code}. Response Text: {response.text}")
|
| 187 |
st.rerun()
|
| 188 |
|
| 189 |
if (st.session_state.messages[-1]["role"] == "assistant") & (st.session_state.messages[-1]["content"] != "How may I help you?"):
|
| 190 |
streamlit_feedback(feedback_type="thumbs", on_submit = thumbs_feedback, key = st.session_state.feedback_key,
|
| 191 |
-
kwargs = {"
|
| 192 |
-
"
|
| 193 |
-
"
|
|
|
|
| 194 |
|
| 195 |
if __name__ == "__main__":
|
| 196 |
launch_bot()
|
|
|
|
| 1 |
from omegaconf import OmegaConf
|
| 2 |
from query import VectaraQuery
|
| 3 |
import os
|
| 4 |
+
from PIL import Image
|
|
|
|
| 5 |
import uuid
|
| 6 |
|
| 7 |
import streamlit as st
|
| 8 |
from streamlit_pills import pills
|
| 9 |
from streamlit_feedback import streamlit_feedback
|
| 10 |
|
| 11 |
+
from utils import thumbs_feedback, send_amplitude_data, escape_dollars_outside_latex
|
| 12 |
+
|
| 13 |
|
| 14 |
max_examples = 6
|
| 15 |
+
languages = {'English': 'eng', 'Spanish': 'spa', 'French': 'fra', 'Chinese': 'zho', 'German': 'deu', 'Hindi': 'hin', 'Arabic': 'ara',
|
| 16 |
'Portuguese': 'por', 'Italian': 'ita', 'Japanese': 'jpn', 'Korean': 'kor', 'Russian': 'rus', 'Turkish': 'tur', 'Persian (Farsi)': 'fas',
|
| 17 |
'Vietnamese': 'vie', 'Thai': 'tha', 'Hebrew': 'heb', 'Dutch': 'nld', 'Indonesian': 'ind', 'Polish': 'pol', 'Ukrainian': 'ukr',
|
| 18 |
'Romanian': 'ron', 'Swedish': 'swe', 'Czech': 'ces', 'Greek': 'ell', 'Bengali': 'ben', 'Malay (or Malaysian)': 'msa', 'Urdu': 'urd'}
|
|
|
|
| 21 |
if 'device_id' not in st.session_state:
|
| 22 |
st.session_state.device_id = str(uuid.uuid4())
|
| 23 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
|
| 25 |
if "feedback_key" not in st.session_state:
|
| 26 |
st.session_state.feedback_key = 0
|
|
|
|
| 31 |
return x.strip().lower() == 'true'
|
| 32 |
|
| 33 |
def launch_bot():
|
| 34 |
+
def reset():
|
| 35 |
+
st.session_state.messages = [{"role": "assistant", "content": "How may I help you?"}]
|
| 36 |
+
st.session_state.ex_prompt = None
|
| 37 |
+
st.session_state.first_turn = True
|
| 38 |
+
|
| 39 |
+
|
| 40 |
def generate_response(question):
|
| 41 |
response = vq.submit_query(question, languages[st.session_state.language])
|
| 42 |
return response
|
|
|
|
| 89 |
cfg.language = st.selectbox('Language:', languages.keys())
|
| 90 |
if st.session_state.language != cfg.language:
|
| 91 |
st.session_state.language = cfg.language
|
| 92 |
+
reset()
|
| 93 |
st.rerun()
|
| 94 |
|
| 95 |
+
st.markdown("\n")
|
| 96 |
+
bc1, _ = st.columns([1, 1])
|
| 97 |
+
with bc1:
|
| 98 |
+
if st.button('Start Over'):
|
| 99 |
+
reset()
|
| 100 |
+
st.rerun()
|
| 101 |
+
|
| 102 |
st.markdown("---")
|
| 103 |
st.markdown(
|
| 104 |
"## How this works?\n"
|
|
|
|
| 112 |
st.markdown(f"<center> <h2> Vectara AI Assistant: {cfg.title} </h2> </center>", unsafe_allow_html=True)
|
| 113 |
|
| 114 |
if "messages" not in st.session_state.keys():
|
| 115 |
+
reset()
|
| 116 |
|
| 117 |
# Display chat messages
|
| 118 |
for message in st.session_state.messages:
|
|
|
|
| 140 |
if st.session_state.messages[-1]["role"] != "assistant":
|
| 141 |
with st.chat_message("assistant"):
|
| 142 |
if cfg.streaming:
|
| 143 |
+
stream = generate_streaming_response(prompt)
|
| 144 |
+
response = st.write_stream(stream)
|
| 145 |
else:
|
| 146 |
with st.spinner("Thinking..."):
|
| 147 |
response = generate_response(prompt)
|
| 148 |
st.write(response)
|
| 149 |
+
|
| 150 |
+
response = escape_dollars_outside_latex(response)
|
| 151 |
message = {"role": "assistant", "content": response}
|
| 152 |
st.session_state.messages.append(message)
|
| 153 |
|
| 154 |
# Send query and response to Amplitude Analytics
|
| 155 |
+
send_amplitude_data(
|
| 156 |
+
user_query=st.session_state.messages[-2]["content"],
|
| 157 |
+
chat_response=st.session_state.messages[-1]["content"],
|
| 158 |
+
demo_name=cfg["title"],
|
| 159 |
+
language=st.session_state.language
|
| 160 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 161 |
st.rerun()
|
| 162 |
|
| 163 |
if (st.session_state.messages[-1]["role"] == "assistant") & (st.session_state.messages[-1]["content"] != "How may I help you?"):
|
| 164 |
streamlit_feedback(feedback_type="thumbs", on_submit = thumbs_feedback, key = st.session_state.feedback_key,
|
| 165 |
+
kwargs = {"user_query": st.session_state.messages[-2]["content"],
|
| 166 |
+
"chat_response": st.session_state.messages[-1]["content"],
|
| 167 |
+
"demo_name": cfg["title"],
|
| 168 |
+
"response_language": st.session_state.language})
|
| 169 |
|
| 170 |
if __name__ == "__main__":
|
| 171 |
launch_bot()
|
utils.py
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import requests
|
| 3 |
+
import json
|
| 4 |
+
import re
|
| 5 |
+
|
| 6 |
+
import streamlit as st
|
| 7 |
+
|
| 8 |
+
headers = {
|
| 9 |
+
'Content-Type': 'application/json',
|
| 10 |
+
'Accept': '*/*'
|
| 11 |
+
}
|
| 12 |
+
|
| 13 |
+
def thumbs_feedback(feedback, **kwargs):
|
| 14 |
+
"""
|
| 15 |
+
Sends feedback to Amplitude Analytics
|
| 16 |
+
"""
|
| 17 |
+
|
| 18 |
+
send_amplitude_data(
|
| 19 |
+
user_query=kwargs.get("user_query", "No user input"),
|
| 20 |
+
chat_response=kwargs.get("chat_response", "No bot response"),
|
| 21 |
+
demo_name=kwargs.get("demo_name", "Unknown"),
|
| 22 |
+
language = kwargs.get("response_language", "Unknown"),
|
| 23 |
+
feedback=feedback["score"],
|
| 24 |
+
)
|
| 25 |
+
st.session_state.feedback_key += 1
|
| 26 |
+
|
| 27 |
+
def send_amplitude_data(user_query, chat_response, demo_name, language, feedback=None):
|
| 28 |
+
# Send query and response to Amplitude Analytics
|
| 29 |
+
data = {
|
| 30 |
+
"api_key": os.getenv('AMPLITUDE_TOKEN'),
|
| 31 |
+
"events": [{
|
| 32 |
+
"device_id": st.session_state.device_id,
|
| 33 |
+
"event_type": "submitted_query",
|
| 34 |
+
"event_properties": {
|
| 35 |
+
"Space Name": demo_name,
|
| 36 |
+
"Demo Type": "chatbot",
|
| 37 |
+
"query": user_query,
|
| 38 |
+
"response": chat_response,
|
| 39 |
+
"Response Language": language
|
| 40 |
+
}
|
| 41 |
+
}]
|
| 42 |
+
}
|
| 43 |
+
if feedback:
|
| 44 |
+
data["events"][0]["event_properties"]["feedback"] = feedback
|
| 45 |
+
|
| 46 |
+
response = requests.post('https://api2.amplitude.com/2/httpapi', headers=headers, data=json.dumps(data))
|
| 47 |
+
if response.status_code != 200:
|
| 48 |
+
print(f"Amplitude request failed with status code {response.status_code}. Response Text: {response.text}")
|
| 49 |
+
|
| 50 |
+
def escape_dollars_outside_latex(text):
|
| 51 |
+
# Define a regex pattern to find LaTeX equations (either single $ or double $$)
|
| 52 |
+
pattern = re.compile(r'(\$\$.*?\$\$|\$.*?\$)')
|
| 53 |
+
latex_matches = pattern.findall(text)
|
| 54 |
+
|
| 55 |
+
# Placeholder to temporarily store LaTeX equations
|
| 56 |
+
placeholders = {}
|
| 57 |
+
for i, match in enumerate(latex_matches):
|
| 58 |
+
placeholder = f'__LATEX_PLACEHOLDER_{i}__'
|
| 59 |
+
placeholders[placeholder] = match
|
| 60 |
+
text = text.replace(match, placeholder)
|
| 61 |
+
|
| 62 |
+
# Escape dollar signs in the rest of the text
|
| 63 |
+
text = text.replace('$', '\\$')
|
| 64 |
+
|
| 65 |
+
# Replace placeholders with the original LaTeX equations
|
| 66 |
+
for placeholder, original in placeholders.items():
|
| 67 |
+
text = text.replace(placeholder, original)
|
| 68 |
+
return text
|