UNTE_ASSISTANT / app.py
tatts's picture
Update app.py
0792b3a verified
raw
history blame
11.3 kB
import os
import re
import logging
import requests
import pandas as pd
from bs4 import BeautifulSoup
from langdetect import detect, DetectorFactory
from langdetect.lang_detect_exception import LangDetectException
import langid
from deep_translator import GoogleTranslator
import gradio as gr
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import Chroma
from langchain.docstore.document import Document
from langchain_community.vectorstores.utils import filter_complex_metadata
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.pydantic_v1 import BaseModel, Field
from langchain_openai import ChatOpenAI
from langchain_core.runnables import RunnablePassthrough, RunnableLambda
from langchain_core.output_parsers import StrOutputParser
from operator import itemgetter
from langchain_community.tools.tavily_search import TavilySearchResults
from typing import List
from typing_extensions import TypedDict
from langgraph.graph import END, StateGraph
from langchain_openai import OpenAIEmbeddings
from langchain_community.document_loaders import UnstructuredURLLoader
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from langchain.memory import ConversationBufferMemory
from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
from langchain.chains import create_history_aware_retriever
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from langchain_core.messages import HumanMessage
# Setup logging
logging.basicConfig(level=logging.DEBUG)
OPENAI_API_TOKEN = "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"
os.environ["OPENAI_API_KEY"] = OPENAI_API_TOKEN
# Retrieve the secret token from environment variables
hf_api_token = os.getenv('HF_API_TOKEN')
# Ensure the token is not None
if hf_api_token is None:
raise ValueError("HF_API_TOKEN environment variable not set")
# Fixing random seed for reproducibility in langdetect
DetectorFactory.seed = 0
# Function to translate text based on detected language
def translate_content(text):
try:
detected_lang = detect(text)
if detected_lang == 'fr':
return GoogleTranslator(source='fr', target='en').translate(text)
elif detected_lang == 'en':
return GoogleTranslator(source='en', target='fr').translate(text)
else:
return text
except Exception as e:
print(f"Error detecting language or translating: {e}")
return text
# Function to chunk content
def chunk_content(content, chunk_size=1250, overlap=250):
chunks = []
start = 0
while start < len(content):
end = start + chunk_size
chunk = content[start:end]
chunks.append(chunk)
start += chunk_size - overlap
return chunks
# Initialize the list to store chunked documents
chunked_web_doc = []
# Load the Excel file
df = pd.read_excel("UNTEanswers.xlsx")
# Merge the 'prompt' and 'reference' columns
df['merged_content'] = df['prompt'] + " " + df['reference']
# Translate and store all text entries in a list
text_entries = []
for index, row in df.iterrows():
# Original content
merged_content = row['merged_content']
text_entries.append(merged_content)
# Translated content
translated_content = translate_content(merged_content)
if translated_content and translated_content != merged_content:
text_entries.append(translated_content)
# Convert the list of text entries into a single string
excel_text = "\n".join(text_entries)
# Process content from the Excel file
for index, row in df.iterrows():
merged_content = row['merged_content']
# Chunk the original content
en_chunks = chunk_content(merged_content)
for chunk in en_chunks:
chunked_web_doc.append({
"url": "UNTEanswers.xlsx", # Mark as coming from the Excel file
"language": detect(merged_content),
"chunk": chunk
})
# Translate and chunk the content if necessary
translated_content = translate_content(merged_content)
if translated_content and translated_content != merged_content:
translated_chunks = chunk_content(translated_content)
for chunk in translated_chunks:
chunked_web_doc.append({
"url": "UNTEanswers.xlsx", # Mark as coming from the Excel file
"language": detect(translated_content),
"chunk": chunk
})
# Load the fetched content from the text file
with open('fetched_contentt.txt', 'r', encoding='utf-8') as f:
fetched_content = f.read()
# Combine the text from the Excel file and the fetched content
content = fetched_content + "\n" + excel_text
# Optionally, save the combined content to a new file
with open('merged_content.txt', 'w', encoding='utf-8') as f:
f.write(content)
web_contents = content.split("-" * 80 + "\n\n")
for block in web_contents:
if block.strip():
lines = block.strip().splitlines()
url = ""
title = ""
en_content = ""
fr_content = ""
language = None
for i, line in enumerate(lines):
if line.startswith("URL:"):
url = line.split("URL:")[1].strip()
elif line.startswith("Title:"):
title = line.split("Title:")[1].strip()
elif line == "English Content:":
language = "en"
elif line == "French Content:":
language = "fr"
else:
if language == "en":
en_content += line + "\n"
elif language == "fr":
fr_content += line + "\n"
if en_content.strip():
en_chunks = chunk_content(en_content.strip())
for chunk in en_chunks:
chunked_web_doc.append({
"url": url,
"language": "en",
"chunk": chunk
})
if fr_content.strip():
fr_chunks = chunk_content(fr_content.strip())
for chunk in fr_chunks:
chunked_web_doc.append({
"url": url,
"language": "fr",
"chunk": chunk
})
model_id = 'sentence-transformers/all-MiniLM-L6-v2'
model_kwargs = {'device': 'cpu'}
embeddings = HuggingFaceEmbeddings(
model_name=model_id,
model_kwargs=model_kwargs
)
documents = [
Document(page_content=chunk['chunk'], metadata={"url": chunk['url'], "language": chunk['language']})
for chunk in chunked_web_doc
]
chroma_db = Chroma.from_documents(documents=documents,
collection_name='rag_web_db',
embedding=embeddings,
collection_metadata={"hnsw:space": "cosine"},
persist_directory="./web_db")
similarity_threshold_retriever = chroma_db.as_retriever(search_type="similarity_score_threshold",
search_kwargs={"k": 3,
"score_threshold": 0.3})
llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
################ history_aware_retriever###################
from langchain.chains import create_history_aware_retriever
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
contextualize_q_system_prompt = """Given a chat history and the latest user question \
which might reference context in the chat history, formulate a standalone question \
which can be understood without the chat history. Do NOT answer the question, \
just reformulate it if needed and otherwise return it as is."""
contextualize_q_prompt = ChatPromptTemplate.from_messages(
[
("system", contextualize_q_system_prompt),
MessagesPlaceholder("chat_history"),
("human", "{input}"),
]
)
history_aware_retriever = create_history_aware_retriever(
llm, similarity_threshold_retriever, contextualize_q_prompt
)
################ question_answer_chain#####################
from langchain.chains import create_retrieval_chain
from langchain.chains.combine_documents import create_stuff_documents_chain
qa_system_prompt = """You are an assistant for question-answering tasks. \
Use the following pieces of retrieved context to answer the question. \
If you don't know the answer, just say that you don't know. \
Use three sentences maximum and keep the answer concise.\
{context}"""
qa_prompt = ChatPromptTemplate.from_messages(
[
("system", qa_system_prompt),
MessagesPlaceholder("chat_history"),
("human", "{input}"),
]
)
question_answer_chain = create_stuff_documents_chain(llm, qa_prompt)
################ rag_chain#####################
rag_chain = create_retrieval_chain(history_aware_retriever, question_answer_chain)
chat_history = []
def ask(question, chat_history):
# Prepend a phrase to the question to ensure relevance to Moodle
prepended_phrase = "using platform Moodle :"
modified_question = prepended_phrase + question
# Invoke the chain to get the response
ai_message = rag_chain.invoke({"input": modified_question, "chat_history": chat_history})
chat_history.append(("user", question))
answer = ai_message["answer"]
# Prepare document links if available
document_links = []
for doc in ai_message.get('context', []):
if 'url' in doc.metadata:
document_links.append(doc.metadata['url'])
# Append the question and answer to the chat history (without sources)
chat_history.append(("assistant", answer))
# For display purposes, format the chat history without labels
display_chat_history = []
for role, content in chat_history:
if role == "user":
display_chat_history.append((None, content)) # User question on the right
else:
display_chat_history.append((content, None)) # Assistant answer on the left
# Add sources to the last assistant message for display purposes only
if document_links:
document_links_text = "\n".join(document_links)
display_chat_history[-1] = (display_chat_history[-1][0] + f"\nSources: {document_links_text}", None)
# Return display history for the UI, and the actual chat history for internal use
return display_chat_history, chat_history, ""
# Initialize the Gradio interface
with gr.Blocks(theme=gr.themes.Soft()) as demo:
chatbot = gr.Chatbot()
clear_button = gr.Button("Clear")
#clear = gr.Button("Clear")
question = gr.Textbox(placeholder="Ask me anything about Moodle...")
chat_history = gr.State([])
question.submit(ask, [question, chat_history], [chatbot, chat_history, question])
clear_button.click(lambda: ([], [], ""), None, [chatbot, chat_history, question], queue=False)
#clear.click(lambda: ("", []), None, [chatbot, chat_history, question], queue=False)
demo.queue()
demo.launch(share=False)