Upload agent.py

agent.py

@@ -1,101 +1,214 @@
-from transformers import Tool, HfAgent
-from huggingface_hub import list_models
-import requests
-from typing import Optional, List
-import random
-
-# First, let's define some custom tools the agent can use
-
-class WebSearchTool(Tool):
-    name = "web_search"
-    description = ("A tool that performs a web search using a search engine API. "
-                   "Input should be a search query. Output will be search results.")
+"""LangGraph Agent"""
+import os
+from dotenv import load_dotenv
+from langgraph.graph import START, StateGraph, MessagesState
+from langgraph.prebuilt import tools_condition
+from langgraph.prebuilt import ToolNode
+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
+from langchain_community.document_loaders import ArxivLoader
+from langchain_community.vectorstores import SupabaseVectorStore
+from langchain_core.messages import SystemMessage, HumanMessage
+from langchain_core.tools import tool
+from langchain.tools.retriever import create_retriever_tool
+from supabase.client import Client, create_client
+
+load_dotenv()
+
+@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.
     
-    inputs = ["text"]
-    outputs = ["text"]
+    Args:
+        a: first int
+        b: second int
+    """
+    return a + b
+
+@tool
+def subtract(a: int, b: int) -> int:
+    """Subtract two numbers.
     
-    def __call__(self, query: str):
-        # In a real implementation, you would call a search API here
-        # For demonstration, we'll return mock results
-        return f"Search results for '{query}': 1. Relevant result 1, 2. Relevant result 2"
-
-class CalculatorTool(Tool):
-    name = "calculator"
-    description = ("A tool for performing mathematical calculations. "
-                   "Input should be a mathematical expression. Output will be the result.")
+    Args:
+        a: first int
+        b: second int
+    """
+    return a - b
+
+@tool
+def divide(a: int, b: int) -> int:
+    """Divide two numbers.
     
-    inputs = ["text"]
-    outputs = ["text"]
+    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.
     
-    def __call__(self, expression: str):
-        try:
-            result = eval(expression)  # Note: In production, use a safer eval method
-            return str(result)
-        except:
-            return "Error: Could not evaluate the expression"
-
-class CurrentTimeTool(Tool):
-    name = "get_current_time"
-    description = ("A tool that returns the current time in UTC. "
-                   "No input needed. Output will be the current time.")
+    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.
     
-    inputs = []
-    outputs = ["text"]
+    Args:
+        query: The search query."""
+    search_docs = WikipediaLoader(query=query, load_max_docs=2).load()
+    formatted_search_docs = "\n\n---\n\n".join(
+        [
+            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
+            for doc in search_docs
+        ])
+    return {"wiki_results": formatted_search_docs}
+
+@tool
+def web_search(query: str) -> str:
+    """Search Tavily for a query and return maximum 3 results.
     
-    def __call__(self):
-        from datetime import datetime
-        return datetime.utcnow().strftime("%Y-%m-%d %H:%M:%S UTC")
-
-class WikipediaTool(Tool):
-    name = "wikipedia_search"
-    description = ("A tool that searches Wikipedia. "
-                   "Input should be a search term. Output will be a summary from Wikipedia.")
+    Args:
+        query: The search query."""
+    search_docs = TavilySearchResults(max_results=3).invoke(query=query)
+    formatted_search_docs = "\n\n---\n\n".join(
+        [
+            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content}\n</Document>'
+            for doc in search_docs
+        ])
+    return {"web_results": formatted_search_docs}
+
+@tool
+def arvix_search(query: str) -> str:
+    """Search Arxiv for a query and return maximum 3 result.
     
-    inputs = ["text"]
-    outputs = ["text"]
+    Args:
+        query: The search query."""
+    search_docs = ArxivLoader(query=query, load_max_docs=3).load()
+    formatted_search_docs = "\n\n---\n\n".join(
+        [
+            f'<Document source="{doc.metadata["source"]}" page="{doc.metadata.get("page", "")}"/>\n{doc.page_content[:1000]}\n</Document>'
+            for doc in search_docs
+        ])
+    return {"arvix_results": formatted_search_docs}
+
+
+
+# load the system prompt from the file
+with open("system_prompt.txt", "r", encoding="utf-8") as f:
+    system_prompt = f.read()
+
+# System message
+sys_msg = SystemMessage(content=system_prompt)
+
+# build a retriever
+embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-mpnet-base-v2") #  dim=768
+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",
+)
+create_retriever_tool = create_retriever_tool(
+    retriever=vector_store.as_retriever(),
+    name="Question Search",
+    description="A tool to retrieve similar questions from a vector store.",
+)
+
+
+
+tools = [
+    multiply,
+    add,
+    subtract,
+    divide,
+    modulus,
+    wiki_search,
+    web_search,
+    arvix_search,
+]
+
+# Build graph function
+def build_graph(provider: str = "groq"):
+    """Build the graph"""
+    # Load environment variables from .env file
+    if provider == "google":
+        # Google Gemini
+        llm = ChatGoogleGenerativeAI(model="gemini-2.0-flash", temperature=0)
+    elif provider == "groq":
+        # Groq https://console.groq.com/docs/models
+        llm = ChatGroq(model="qwen-qwq-32b", temperature=0) # optional : qwen-qwq-32b gemma2-9b-it
+    elif provider == "huggingface":
+        # TODO: Add huggingface endpoint
+        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)
+
+    # Node
+    def assistant(state: MessagesState):
+        """Assistant node"""
+        return {"messages": [llm_with_tools.invoke(state["messages"])]}
     
-    def __call__(self, term: str):
-        # Mock implementation - in real use, you'd call the Wikipedia API
-        return f"Wikipedia summary for '{term}': This is a summary about {term}."
-
-# Now let's create the agent with these tools and some pre-trained tools
-
-def create_agent():
-    # Load pre-trained tools from the Hub
-    agent = HfAgent(
-        "https://api-inference.huggingface.co/models/bigcode/starcoder",
-        additional_tools=[
-            WebSearchTool(),
-            CalculatorTool(),
-            CurrentTimeTool(),
-            WikipediaTool()
-        ],
-        # These parameters help with performance
-        max_new_tokens=200,
-        temperature=0.7,
-        top_p=0.9,
+    def retriever(state: MessagesState):
+        """Retriever node"""
+        similar_question = vector_store.similarity_search(state["messages"][0].content)
+        example_msg = HumanMessage(
+            content=f"Here I provide a similar question and answer for reference: \n\n{similar_question[0].page_content}",
+        )
+        return {"messages": [sys_msg] + state["messages"] + [example_msg]}
+
+    builder = StateGraph(MessagesState)
+    builder.add_node("retriever", retriever)
+    builder.add_node("assistant", assistant)
+    builder.add_node("tools", ToolNode(tools))
+    builder.add_edge(START, "retriever")
+    builder.add_edge("retriever", "assistant")
+    builder.add_conditional_edges(
+        "assistant",
+        tools_condition,
     )
-    return agent
+    builder.add_edge("tools", "assistant")
+
+    # Compile graph
+    return builder.compile()
 
-# Example usage of the agent
+# test
 if __name__ == "__main__":
-    agent = create_agent()
-    
-    # Test the agent with some sample queries
-    queries = [
-        "What's the capital of France?",
-        "Calculate 123 * 45",
-        "What time is it now?",
-        "Tell me about Albert Einstein",
-        "Search for the latest news about AI"
-    ]
-    
-    for query in queries:
-        print(f"Query: {query}")
-        result = agent.run(query)
-        print(f"Result: {result}\n")
-    
-    # To evaluate on the benchmark, you would use:
-    # from transformers.benchmarks import evaluate_agent
-    # benchmark_score = evaluate_agent(agent)
-    # print(f"Benchmark score: {benchmark_score}")
+    question = "When was a picture of St. Thomas Aquinas first added to the Wikipedia page on the Principle of double effect?"
+    # Build the graph
+    graph = build_graph(provider="groq")
+    # Run the graph
+    messages = [HumanMessage(content=question)]
+    messages = graph.invoke({"messages": messages})
+    for m in messages["messages"]:
+        m.pretty_print()