josondev's picture
Update veryfinal.py
ca98093 verified
raw
history blame
12.1 kB
"""
Enhanced LangGraph Agent with Multi-LLM Support and Proper Question Answering
Combines your original LangGraph structure with enhanced response handling
"""
import os
import time
import random
from dotenv import load_dotenv
from typing import List, Dict, Any, TypedDict, Annotated
import operator
from langgraph.graph import START, StateGraph, MessagesState, END
from langgraph.prebuilt import tools_condition, ToolNode
from langgraph.checkpoint.memory import MemorySaver
from langchain_google_genai import ChatGoogleGenerativeAI
from langchain_groq import ChatGroq
from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint, HuggingFaceEmbeddings
from langchain_community.tools.tavily_search import TavilySearchResults
from langchain_community.document_loaders import WikipediaLoader, ArxivLoader
from langchain_community.vectorstores import SupabaseVectorStore
from langchain_core.messages import SystemMessage, HumanMessage, AIMessage
from langchain_core.tools import tool
from langchain.tools.retriever import create_retriever_tool
from supabase.client import Client, create_client
load_dotenv()
# Enhanced system prompt for better question answering
ENHANCED_SYSTEM_PROMPT = """You are a helpful assistant tasked with answering questions using a set of tools.
CRITICAL INSTRUCTIONS:
1. Read the question carefully and understand what specific information is being asked
2. Use the appropriate tools to find the exact information requested
3. For factual questions, search for current and accurate information
4. For calculations, use the math tools provided
5. Always provide specific, direct answers - never repeat the question as your answer
6. If you cannot find the information, state "Information not available"
7. Format your final response as: FINAL ANSWER: [your specific answer]
ANSWER FORMAT RULES:
- For numbers: provide just the number without commas or units unless specified
- For names/strings: provide the exact name or term without articles
- For lists: provide comma-separated values
- Be concise and specific in your final answer
Remember: Your job is to ANSWER the question, not repeat it back."""
# ---- Enhanced Tool Definitions ----
@tool
def multiply(a: int, b: int) -> int:
"""Multiply two numbers.
Args:
a: first int
b: second int
"""
return a * b
@tool
def add(a: int, b: int) -> int:
"""Add two numbers.
Args:
a: first int
b: second int
"""
return a + b
@tool
def subtract(a: int, b: int) -> int:
"""Subtract two numbers.
Args:
a: first int
b: second int
"""
return a - b
@tool
def divide(a: int, b: int) -> float:
"""Divide two numbers.
Args:
a: first int
b: second int
"""
if b == 0:
raise ValueError("Cannot divide by zero.")
return a / b
@tool
def modulus(a: int, b: int) -> int:
"""Get the modulus of two numbers.
Args:
a: first int
b: second int
"""
return a % b
@tool
def wiki_search(query: str) -> str:
"""Search Wikipedia for a query and return maximum 2 results.
Args:
query: The search query.
"""
try:
time.sleep(random.uniform(0.5, 1.0)) # Rate limiting
search_docs = WikipediaLoader(query=query, load_max_docs=2).load()
if not search_docs:
return "No Wikipedia results found"
formatted_search_docs = "\n\n---\n\n".join([
f'<Document source="{doc.metadata.get("source", "Wikipedia")}" title="{doc.metadata.get("title", "")}">\n{doc.page_content[:1500]}\n</Document>'
for doc in search_docs
])
return formatted_search_docs
except Exception as e:
return f"Wikipedia search failed: {e}"
@tool
def web_search(query: str) -> str:
"""Search Tavily for a query and return maximum 3 results.
Args:
query: The search query.
"""
try:
time.sleep(random.uniform(0.7, 1.2)) # Rate limiting
search_tool = TavilySearchResults(max_results=3)
search_docs = search_tool.invoke({"query": query})
if not search_docs:
return "No web search results found"
formatted_search_docs = "\n\n---\n\n".join([
f'<Document source="{doc.get("url", "")}">\n{doc.get("content", "")[:1200]}\n</Document>'
for doc in search_docs
])
return formatted_search_docs
except Exception as e:
return f"Web search failed: {e}"
@tool
def arxiv_search(query: str) -> str:
"""Search Arxiv for a query and return maximum 3 results.
Args:
query: The search query.
"""
try:
time.sleep(random.uniform(0.5, 1.0)) # Rate limiting
search_docs = ArxivLoader(query=query, load_max_docs=3).load()
if not search_docs:
return "No ArXiv results found"
formatted_search_docs = "\n\n---\n\n".join([
f'<Document source="{doc.metadata.get("source", "ArXiv")}" title="{doc.metadata.get("title", "")}">\n{doc.page_content[:1000]}\n</Document>'
for doc in search_docs
])
return formatted_search_docs
except Exception as e:
return f"ArXiv search failed: {e}"
# Initialize tools list
tools = [
multiply, add, subtract, divide, modulus,
wiki_search, web_search, arxiv_search
]
# Enhanced State for better tracking
class EnhancedState(MessagesState):
"""Enhanced state with additional tracking"""
query: str = ""
tools_used: List[str] = []
search_results: str = ""
def build_graph(provider: str = "groq"):
"""Build the enhanced graph with proper error handling and response formatting"""
# Initialize LLM based on provider
if provider == "google":
llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)
elif provider == "groq":
llm = ChatGroq(model="llama3-70b-8192", temperature=0) # Using more reliable model
elif provider == "huggingface":
llm = ChatHuggingFace(
llm=HuggingFaceEndpoint(
url="https://api-inference.huggingface.co/models/Meta-DeepLearning/llama-2-7b-chat-hf",
temperature=0,
),
)
else:
raise ValueError("Invalid provider. Choose 'google', 'groq' or 'huggingface'.")
# Bind tools to LLM
llm_with_tools = llm.bind_tools(tools)
# Initialize vector store if available
vector_store = None
try:
if os.getenv("SUPABASE_URL") and os.getenv("SUPABASE_SERVICE_KEY"):
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2")
supabase: Client = create_client(
os.environ.get("SUPABASE_URL"),
os.environ.get("SUPABASE_SERVICE_KEY")
)
vector_store = SupabaseVectorStore(
client=supabase,
embedding=embeddings,
table_name="documents",
query_name="match_documents_langchain",
)
except Exception as e:
print(f"Vector store initialization failed: {e}")
def retriever(state: MessagesState):
"""Enhanced retriever node with fallback"""
messages = state["messages"]
query = messages[-1].content if messages else ""
# Try to get similar questions from vector store
similar_context = ""
if vector_store:
try:
similar_questions = vector_store.similarity_search(query, k=1)
if similar_questions:
similar_context = f"\n\nSimilar example for reference:\n{similar_questions[0].page_content}"
except Exception as e:
print(f"Vector search failed: {e}")
# Enhanced system message with context
enhanced_prompt = ENHANCED_SYSTEM_PROMPT + similar_context
sys_msg = SystemMessage(content=enhanced_prompt)
return {"messages": [sys_msg] + messages}
def assistant(state: MessagesState):
"""Enhanced assistant node with better response handling"""
try:
response = llm_with_tools.invoke(state["messages"])
# Ensure response is properly formatted
if hasattr(response, 'content'):
content = response.content
# Check if this is just repeating the question
original_query = state["messages"][-1].content if state["messages"] else ""
if content.strip() == original_query.strip():
# Force a better response
enhanced_messages = state["messages"] + [
HumanMessage(content=f"Please provide a specific answer to this question, do not repeat the question: {original_query}")
]
response = llm_with_tools.invoke(enhanced_messages)
return {"messages": [response]}
except Exception as e:
error_response = AIMessage(content=f"Error processing request: {e}")
return {"messages": [error_response]}
def format_final_answer(state: MessagesState):
"""Format the final answer properly"""
messages = state["messages"]
if not messages:
return {"messages": [AIMessage(content="FINAL ANSWER: Information not available")]}
last_message = messages[-1]
if hasattr(last_message, 'content'):
content = last_message.content
# Ensure proper formatting
if "FINAL ANSWER:" not in content:
# Extract the key information and format it
if content.strip():
formatted_content = f"FINAL ANSWER: {content.strip()}"
else:
formatted_content = "FINAL ANSWER: Information not available"
formatted_message = AIMessage(content=formatted_content)
return {"messages": messages[:-1] + [formatted_message]}
return {"messages": messages}
# Build the graph
builder = StateGraph(MessagesState)
# Add nodes
builder.add_node("retriever", retriever)
builder.add_node("assistant", assistant)
builder.add_node("tools", ToolNode(tools))
builder.add_node("formatter", format_final_answer)
# Add edges
builder.add_edge(START, "retriever")
builder.add_edge("retriever", "assistant")
builder.add_conditional_edges(
"assistant",
tools_condition,
{
"tools": "tools",
"__end__": "formatter"
}
)
builder.add_edge("tools", "assistant")
builder.add_edge("formatter", END)
# Compile graph with checkpointer
return builder.compile(checkpointer=MemorySaver())
# Test function
def test_agent():
"""Test the agent with sample questions"""
graph = build_graph(provider="groq")
test_questions = [
"How many studio albums were published by Mercedes Sosa between 2000 and 2009?",
"What is 25 multiplied by 17?",
"Who nominated the only Featured Article on English Wikipedia about a dinosaur that was promoted in November 2004?"
]
for question in test_questions:
print(f"\nQuestion: {question}")
print("-" * 60)
try:
messages = [HumanMessage(content=question)]
config = {"configurable": {"thread_id": f"test_{hash(question)}"}}
result = graph.invoke({"messages": messages}, config)
if result and "messages" in result:
final_message = result["messages"][-1]
if hasattr(final_message, 'content'):
print(f"Answer: {final_message.content}")
else:
print(f"Answer: {final_message}")
else:
print("Answer: No response generated")
except Exception as e:
print(f"Error: {e}")
print()
if __name__ == "__main__":
# Run tests
test_agent()