update by using smolagent

app.py

@@ -4,8 +4,7 @@ import gradio as gr
 import requests
 import inspect
 import pandas as pd
-from agent import SmoalAgent
-from tools import search_tool, rag_chain, extract_final_answer
+from tools import search_tool, rag_chain, extract_final_answer, initialize_code_agent
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
@@ -13,7 +12,7 @@ SUBMISSION_FILE = "submission.jsonl"
 
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
-    """Fetches all questions, runs the SmoalAgent on them, submits all answers,
+    """Fetches all questions, runs the CodeAgent on them, submits all answers,
     and displays the results."""
     # --- Determine HF Space Runtime URL and Repo URL ---
     space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
@@ -31,7 +30,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     # 1. Instantiate Agent
     try:
-        agent = SmoalAgent()
+        agent = initialize_code_agent()
+        if not agent:
+            raise Exception("Failed to initialize CodeAgent")
     except Exception as e:
         print(f"Error instantiating agent: {e}")
         return f"Error initializing agent: {e}", None
@@ -71,7 +72,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            # 使用导入的搜索工具和RAG链
+            # Use imported search tool and RAG chain
             search_result = search_tool.run(question_text)
             if rag_chain:
                 response = rag_chain.run(f"{question_text}\nSearch result: {search_result}")
@@ -83,7 +84,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
                     "reasoning_trace": response
                 })
             else:
-                submitted_answer = agent(question_text)
+                submitted_answer = agent.run(question_text)
                 answers_payload.append({
                     "task_id": task_id,
                     "model_answer": submitted_answer
@@ -163,7 +164,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
 # --- Build Gradio Interface using Blocks ---
 with gr.Blocks() as demo:
-    gr.Markdown("# Smoal Agent Evaluation Runner")
+    gr.Markdown("# Code Agent Evaluation Runner")
     gr.Markdown(
         """
         **Instructions:**
@@ -213,5 +214,5 @@ if __name__ == "__main__":
 
     print("-"*(60 + len(" App Starting ")) + "\n")
 
-    print("Launching Gradio Interface for Smoal Agent Evaluation...")
+    print("Launching Gradio Interface for Code Agent Evaluation...")
     demo.launch(debug=True, share=False)

tools.py

@@ -7,6 +7,7 @@ from langchain.vectorstores import FAISS
 from langchain.prompts import PromptTemplate
 from datasets import load_dataset
 from agent import SmoalAgent
+from smolagents import CodeAgent, DuckDuckGoSearchTool, InferenceClientModel
 
 # System prompt for formatting answers
 SYSTEM_PROMPT = """
@@ -58,4 +59,89 @@ def extract_final_answer(response):
 
 # Initialize RAG chain
 global rag_chain
-rag_chain = load_gaia_and_setup_rag()
+rag_chain = load_gaia_and_setup_rag()
+
+# Initialize search tool
+search_tool = DuckDuckGoSearchTool()
+
+# Load GAIA dataset and setup RAG
+rag_chain = None
+
+def load_gaia_and_setup_rag():
+    try:
+        from datasets import load_dataset
+        # Load GAIA dataset (test split)
+        dataset = load_dataset("gaia-benchmark/gaia", split="test")
+        
+        # Extract contexts from dataset
+        contexts = [item["context"] for item in dataset if "context" in item and item["context"]]
+        
+        # Create embeddings and vector store
+        embeddings = OpenAIEmbeddings()
+        vector_store = FAISS.from_texts(contexts, embeddings)
+        
+        # Create retriever
+        retriever = vector_store.as_retriever(search_kwargs={"k": 3})
+        
+        # Define prompt template
+        SYSTEM_PROMPT = """
+        You are a precise QA system. Answer ONLY with the exact answer, no explanations.
+        Answers must be in one of these formats:
+        - A single number
+        - A single string
+        - A comma-separated list of numbers or strings
+        Do not include any additional text, explanations, or formatting.
+        """
+        
+        prompt_template = PromptTemplate(
+            template=SYSTEM_PROMPT + "\nContext: {context}\nQuestion: {question}\nAnswer:",
+            input_variables=["context", "question"]
+        )
+        
+        # Create RAG chain
+        global rag_chain
+        rag_chain = RetrievalQA.from_chain_type(
+            llm=OpenAI(temperature=0),
+            chain_type="stuff",
+            retriever=retriever,
+            chain_type_kwargs={"prompt": prompt_template}
+        )
+        
+        print(f"Successfully loaded GAIA dataset and created RAG chain with {len(contexts)} contexts")
+        return True
+    except Exception as e:
+        print(f"Error setting up RAG: {e}")
+        return False
+
+# Initialize RAG when the module is loaded
+load_gaia_and_setup_rag()
+
+# Initialize CodeAgent
+def initialize_code_agent():
+    try:
+        # Initialize model with environment variables
+        model = InferenceClientModel(
+            api_key=os.getenv("OPENAI_API_KEY"),
+            model_name="gpt-3.5-turbo"
+        )
+        
+        # Create agent with search tool
+        agent = CodeAgent(
+            tools=[search_tool],
+            model=model
+        )
+        
+        print("CodeAgent initialized successfully")
+        return agent
+    except Exception as e:
+        print(f"Error initializing CodeAgent: {e}")
+        return None
+
+# Final answer extraction
+def extract_final_answer(text):
+    # Use regex to find the final answer pattern
+    match = re.search(r'FINAL ANSWER: (.*)', text, re.IGNORECASE)
+    if match:
+        return match.group(1).strip()
+    # If no pattern found, return the text as is (with cleanup)
+    return text.strip()