Update app.py

app.py

@@ -4,12 +4,13 @@ import requests
 import pandas as pd
 from typing import TypedDict, Annotated, Sequence
 import operator
-from langchain_core.messages import BaseMessage, HumanMessage, ToolMessage
+from langchain_core.messages import BaseMessage, HumanMessage
 from langchain.agents import AgentExecutor
-from langchain_community.tools.tavily_search import TavilySearchResults
+from langchain_community.tools import DuckDuckGoSearchRun
 from langchain_huggingface import HuggingFaceEndpoint
 from langgraph.graph import StateGraph, END
 from langgraph.prebuilt import ToolNode, tools_condition
+from langchain_core.prompts import ChatPromptTemplate
 
 # --- Main Application Logic ---
 
@@ -21,28 +22,27 @@ class AgentState(TypedDict):
 def create_langgraph_agent():
     print("Initializing LangGraph Agent...")
     
-    # 1. Set up the LLM (The "Brain")
-    # We use the powerful Mistral model with the correct 'conversational' task
+    # 1. Set up the LLM (The "Brain") using the DeepSeek Coder model
     llm = HuggingFaceEndpoint(
-        repo_id="mistralai/Mistral-7B-Instruct-v0.2",
+        repo_id="deepseek-ai/deepseek-coder-6.7b-instruct",
         task="conversational",
-        max_new_tokens=512,
+        max_new_tokens=1024, # Increased tokens for better reasoning
         do_sample=False,
     )
-    # We bind the tools to the LLM so it knows about them
-    tools = [TavilySearchResults(max_results=3)]
-    llm_with_tools = llm.bind_tools(tools)
+    
+    # We use a wrapper to make the model compatible with LangChain's tool calling
+    from langchain_huggingface.chat_models import HuggingFaceChat
+    llm_with_tools = HuggingFaceChat(endpoint=llm).bind_tools([DuckDuckGoSearchRun()])
     print("LLM and tools initialized.")
 
-    # 2. Define the Graph Nodes
-    # The 'agent' node calls the LLM
+    # 2. Define the agent's logic (the "agent" node)
+    # We define the graph nodes and edges for the agent's reasoning process
     def agent_node(state):
         print("Calling agent node...")
         response = llm_with_tools.invoke(state["messages"])
         return {"messages": [response]}
     
-    # The 'tool' node executes the tools
-    tool_node = ToolNode(tools)
+    tool_node = ToolNode([DuckDuckGoSearchRun()])
     print("Graph nodes defined.")
 
     # 3. Define the Graph
@@ -51,12 +51,7 @@ def create_langgraph_agent():
     graph.add_node("tools", tool_node)
     
     graph.set_entry_point("agent")
-    
-    # This conditional edge decides whether to call a tool or end
-    graph.add_conditional_edges(
-        "agent",
-        tools_condition,
-    )
+    graph.add_conditional_edges("agent", tools_condition)
     graph.add_edge("tools", "agent")
 
     # 4. Compile the graph into a runnable app
@@ -125,7 +120,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
 # Gradio Interface
 with gr.Blocks() as demo:
-    gr.Markdown("# Agent Evaluation Runner (Final LangGraph v2)")
+    gr.Markdown("# Agent Evaluation Runner (DeepSeek + DuckDuckGo)")
     gr.LoginButton()
     run_button = gr.Button("Run Evaluation & Submit All Answers")
     status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)