Upload app.py

app.py

@@ -23,13 +23,104 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # --- Basic Agent Definition ---
 class BasicAgent:
+    """A LlamaIndex-based agent."""
     def __init__(self):
         print("BasicAgent initialized.")
+        try:
+            # Initialize the core components
+            self.llm = self._initialize_llm()
+            
+            # Import get_tools from agent.py here to avoid circular imports
+            from agent import get_tools
+            self.tools = get_tools()
+            
+            self.memory = ChatMemoryBuffer.from_defaults(token_limit=3900)
+            # Build the agent
+            self.agent = self._build_agent()
+            print("Agent setup complete.")
+        except Exception as e:
+            print(f"Warning: Error during agent initialization: {e}")
+            # Continue despite error - we'll handle this in the __call__ method
+        
+    def _initialize_llm(self) -> LLM:
+        """Initialize the LLM based on configuration."""
+        provider = os.getenv("DEFAULT_LLM_PROVIDER", "gemini").lower()
+        
+        if provider == "gemini":
+            api_key = os.getenv("GOOGLE_API_KEY")
+            if not api_key:
+                raise ValueError("GOOGLE_API_KEY not found in environment variables")
+            
+            return Gemini(
+                model_name="models/gemini-1.5-flash",
+                api_key=api_key,
+                temperature=0.1,
+                top_p=0.95,
+                max_tokens=1024,
+            )
+            
+        elif provider == "huggingface":
+            api_key = os.getenv("HUGGINGFACE_API_KEY")
+            if not api_key:
+                raise ValueError("HUGGINGFACE_API_KEY not found in environment variables")
+            
+            return HuggingFaceInferenceAPI(
+                model_name="Qwen/Qwen2.5-Coder-32B-Instruct",
+                api_key=api_key,
+                temperature=0.1,
+                max_tokens=1024,
+            )
+        else:
+            raise ValueError(f"Unsupported LLM provider: {provider}")
+    
+    def _build_agent(self) -> ReActAgent:
+        """Build and return the agent."""
+        # Load system prompt from file and append output format requirements
+        try:
+            with open("system_prompt.txt", "r", encoding="utf-8") as f:
+                system_prompt = f.read()
+            # Append output format to system prompt
+            system_prompt = f"{system_prompt}\n\nIMPORTANT OUTPUT FORMAT:\n{OUTPUT_FORMAT}"
+        except Exception as e:
+            print(f"Error loading system prompt: {e}")
+            system_prompt = f"You are an intelligent agent designed to answer a wide variety of questions.\n\nIMPORTANT OUTPUT FORMAT:\n{OUTPUT_FORMAT}"
+        
+        return ReActAgent.from_tools(
+            tools=self.tools,
+            llm=self.llm,
+            memory=self.memory,
+            system_prompt=system_prompt,
+            verbose=True,
+        )
+
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
-        fixed_answer = "This is a default answer."
-        print(f"Agent returning fixed answer: {fixed_answer}")
-        return fixed_answer
+        
+        try:
+            # Check if agent was properly initialized
+            if not hasattr(self, 'agent') or self.agent is None:
+                # Fallback to a simple response if agent initialization failed
+                return "I'm unable to process your request due to initialization errors."
+                
+            # Process the question
+            response = self.agent.query(question)
+            answer_text = str(response)
+            
+            # Extract the FINAL ANSWER part if it exists
+            if "FINAL ANSWER:" in answer_text:
+                reasoning_trace = answer_text.split("FINAL ANSWER:")[0].strip()
+                model_answer = answer_text.split("FINAL ANSWER:")[1].strip()
+                
+                print(f"Agent generated answer: {model_answer[:50]}..." if len(model_answer) > 50 else f"Agent generated answer: {model_answer}")
+                return model_answer  # Return just the answer part
+            else:
+                # If no FINAL ANSWER pattern, return the whole response
+                print(f"No 'FINAL ANSWER' found in response. Returning full response.")
+                return answer_text
+                
+        except Exception as e:
+            print(f"Error generating answer: {e}")
+            return f"I encountered an error while answering your question: {str(e)}"
 
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     """
@@ -95,15 +186,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            # Get agent response which is now a JSON string
-            agent_response_json = agent(question_text)
-            agent_response = json.loads(agent_response_json)
-            
-            model_answer = agent_response.get("model_answer", "")
-            reasoning_trace = agent_response.get("reasoning_trace", "")
-            
-            # Format for submission payload
-            submitted_answer = model_answer
+            # Get agent response - now it's a direct string
+            submitted_answer = agent(question_text)
             
             # Add to answers payload
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
@@ -112,15 +196,14 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
             results_log.append({
                 "Task ID": task_id, 
                 "Question": question_text, 
-                "Submitted Answer": submitted_answer,
-                "Reasoning": reasoning_trace[:100] + "..." if len(reasoning_trace) > 100 else reasoning_trace
+                "Submitted Answer": submitted_answer
             })
             
-            # Add to JSONL output
+            # Add to JSONL output - save both the answer and reasoning
             jsonl_output.append({
                 "task_id": task_id,
-                "model_answer": model_answer,
-                "reasoning_trace": reasoning_trace
+                "model_answer": submitted_answer,
+                "reasoning_trace": ""  # No separate reasoning trace now
             })
             
         except Exception as e: