Spaces:
Sleeping
Sleeping
Create chat_bot.py
Browse files- chat_bot.py +109 -0
chat_bot.py
ADDED
|
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from openai import OpenAI
|
| 2 |
+
import gradio as gr
|
| 3 |
+
import requests
|
| 4 |
+
from datetime import date
|
| 5 |
+
from test_web_rag import get_docs_from_web
|
| 6 |
+
import json
|
| 7 |
+
import os
|
| 8 |
+
from dotenv import load_dotenv
|
| 9 |
+
|
| 10 |
+
load_dotenv()
|
| 11 |
+
# Replace with your key
|
| 12 |
+
client = OpenAI()
|
| 13 |
+
you_key = os.getenv("YOU_API_KEY")
|
| 14 |
+
|
| 15 |
+
|
| 16 |
+
def get_ai_snippets_for_query(query):
|
| 17 |
+
headers = {"X-API-Key": you_key}
|
| 18 |
+
params = {"query": query}
|
| 19 |
+
return requests.get(
|
| 20 |
+
f"https://api.ydc-index.io/search?query={query}",
|
| 21 |
+
params=params,
|
| 22 |
+
headers=headers,
|
| 23 |
+
).json().get('hits')
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
def get_web_search_you(query):
|
| 27 |
+
docs = get_ai_snippets_for_query(query)
|
| 28 |
+
markdown = ""
|
| 29 |
+
for doc in docs:
|
| 30 |
+
for key, value in doc.items():
|
| 31 |
+
if key == 'snippets':
|
| 32 |
+
markdown += f"{key}:\n"
|
| 33 |
+
for snippet in value:
|
| 34 |
+
markdown += f"- {snippet}\n"
|
| 35 |
+
else:
|
| 36 |
+
markdown += f"{key}: {value}\n"
|
| 37 |
+
markdown += "\n"
|
| 38 |
+
return markdown
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
def predict(message, history, _n_web_search, _strategy):
|
| 42 |
+
# docs = get_web_search_you(message)
|
| 43 |
+
|
| 44 |
+
with open('history.json', mode='a', encoding='utf-8') as f:
|
| 45 |
+
json.dump(history, f)
|
| 46 |
+
docs = get_docs_from_web(message, history[-1:], _n_web_search, _strategy)
|
| 47 |
+
partial_message = ''
|
| 48 |
+
information = ''
|
| 49 |
+
for doc in docs:
|
| 50 |
+
if isinstance(doc, dict):
|
| 51 |
+
information = doc.get('data')
|
| 52 |
+
else:
|
| 53 |
+
partial_message = partial_message + doc
|
| 54 |
+
yield partial_message
|
| 55 |
+
system_prompt = """
|
| 56 |
+
You are an advanced chatbot.
|
| 57 |
+
Today's date - {date}
|
| 58 |
+
|
| 59 |
+
When answering a question, adhere to the following revised rules:
|
| 60 |
+
- The "Information for reference" data is provided in the chunks with each chunk having its own source as url.
|
| 61 |
+
- Generate human-like text in response to input, reflecting your status as a sophisticated language model.
|
| 62 |
+
- Abstain from offering any health or medical advice and ensure all responses maintain this guideline strictly.
|
| 63 |
+
- Format all responses in markdown format consistently throughout interactions.
|
| 64 |
+
- Must cite sources from the information at the conclusion of your response using properly titled references, but only if the information you provided comes from sources that can be cited.
|
| 65 |
+
|
| 66 |
+
Information for reference:
|
| 67 |
+
"{context}"
|
| 68 |
+
|
| 69 |
+
Your answer should be structured in markdown as follows:
|
| 70 |
+
|
| 71 |
+
|
| 72 |
+
<Answer>
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
**Sources**:
|
| 76 |
+
Include this section only if the provided information contains sources. If sources are included, list them as follows:
|
| 77 |
+
- [Title of Source 1](URL to Source 1)
|
| 78 |
+
- [Title of Source 2](URL to Source 2)
|
| 79 |
+
... as needed. If no sources are provided, do not include this section in answer.
|
| 80 |
+
""".format(context=information, question=message, date=date.today().strftime('%B %d, %Y'))
|
| 81 |
+
|
| 82 |
+
history_openai_format = [{"role": "system", "content": system_prompt}]
|
| 83 |
+
for human, assistant in history:
|
| 84 |
+
history_openai_format.append({"role": "user", "content": human})
|
| 85 |
+
history_openai_format.append({"role": "assistant", "content": assistant})
|
| 86 |
+
history_openai_format.append({"role": "user", "content": message})
|
| 87 |
+
# print(history_openai_format)
|
| 88 |
+
|
| 89 |
+
response = client.chat.completions.create(model='gpt-4-turbo',
|
| 90 |
+
messages=history_openai_format,
|
| 91 |
+
temperature=0.5,
|
| 92 |
+
max_tokens=1000,
|
| 93 |
+
top_p=0.5,
|
| 94 |
+
stream=True)
|
| 95 |
+
partial_message += '\n\n'
|
| 96 |
+
for chunk in response:
|
| 97 |
+
if chunk.choices[0].delta.content is not None:
|
| 98 |
+
partial_message = partial_message + chunk.choices[0].delta.content
|
| 99 |
+
yield partial_message
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
n_web_search = gr.Slider(1, 10, value=3, step=1, label="Web searches",
|
| 103 |
+
info="Choose between 1 and 10 number of web searches to do. Remember more the web searches more it will take time to reply.")
|
| 104 |
+
strategy = gr.Radio(["Deep", "Normal"], label="Strategy", value="Normal",
|
| 105 |
+
info="Select web search analysis type. Please keep in mind that deep analysis will take more time than normal analysis.")
|
| 106 |
+
|
| 107 |
+
app = gr.ChatInterface(predict, additional_inputs=[n_web_search, strategy])
|
| 108 |
+
app.queue(default_concurrency_limit=5)
|
| 109 |
+
app.launch(debug=True, share=False)
|