Update app.py

app.py

@@ -1,194 +1,307 @@
 import os
+import re
 import json
-import time
 import requests
-import gradio as gr
-
-# ===============================
-# 1. LLM Wrapper (Your Original)
-# ===============================
-class OpenRouterLLM:
-    def __init__(self, api_key=None, model="deepseek/deepseek-v3.1-terminus"):
-        self.api_key = api_key or os.getenv("OPENROUTER_API_KEY")
-        self.model = model
-        self.base_url = "https://openrouter.ai/api/v1"
-        if not self.api_key:
-            raise ValueError("Missing OpenRouter API key. Set OPENROUTER_API_KEY environment variable.")
+import pandas as pd
+from pathlib import Path
+from typing import Optional, Union, Dict, Any, List
+from dotenv import load_dotenv
+
+from langgraph.graph import StateGraph, MessagesState
+from langgraph.prebuilt import create_react_agent
+from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_core.tools import tool
+from langchain_openai import ChatOpenAI
+
+load_dotenv()
+
+
+class OpenRouterLLM(ChatOpenAI):
+    """Custom OpenRouter LLM wrapper for LangGraph"""
     
-    def generate(self, prompt, system_prompt="You are a helpful AI agent."):
-        headers = {
-            "Authorization": f"Bearer {self.api_key}",
-            "Content-Type": "application/json",
-        }
-        payload = {
-            "model": self.model,
-            "messages": [
-                {"role": "system", "content": system_prompt},
-                {"role": "user", "content": prompt}
-            ],
-            "temperature": 0.1,
-            "max_tokens": 500
-        }
-        try:
-            response = requests.post(f"{self.base_url}/chat/completions", headers=headers, json=payload)
-            response.raise_for_status()
-            data = response.json()
-            return data["choices"][0]["message"]["content"].strip()
-        except Exception as e:
-            return f"Error: {e}"
+    def __init__(self, model: str = "deepseek/deepseek-v3.1-terminus", **kwargs):
+        api_key = os.getenv("OPENROUTER_API_KEY") or os.getenv("my_key")
+        
+        super().__init__(
+            model=model,
+            openai_api_key=api_key,
+            openai_api_base="https://openrouter.ai/api/v1",
+            **kwargs
+        )
 
-# ===============================
-# 2. GAIA API Loader
-# ===============================
-GAIA_API_BASE = "https://gaia-benchmark-hf.fly.dev"
 
-class GAIAAgent:
-    def __init__(self, llm: OpenRouterLLM):
-        self.llm = llm
-        self.questions = []
+@tool
+def search_web(query: str) -> str:
+    """Search the web using DuckDuckGo for current information."""
+    try:
+        # Simple web search using DuckDuckGo
+        search_url = f"https://api.duckduckgo.com/?q={query}&format=json&no_html=1&skip_disambig=1"
+        response = requests.get(search_url, timeout=10)
+        
+        if response.status_code == 200:
+            data = response.json()
+            
+            # Extract results
+            results = []
+            if data.get("AbstractText"):
+                results.append(f"Abstract: {data['AbstractText']}")
+            
+            if data.get("RelatedTopics"):
+                for topic in data["RelatedTopics"][:3]:
+                    if isinstance(topic, dict) and topic.get("Text"):
+                        results.append(f"Related: {topic['Text']}")
+            
+            if results:
+                return "\n".join(results)
+            else:
+                return f"Search performed for '{query}' but no specific results found."
+        else:
+            return f"Search failed with status code {response.status_code}"
+            
+    except Exception as e:
+        return f"Search error: {str(e)}"
+
 
-    def fetch_questions(self):
-        try:
-            resp = requests.get(f"{GAIA_API_BASE}/questions", timeout=30)
-            if resp.status_code == 200:
-                self.questions = resp.json()
-                return f"✅ Loaded {len(self.questions)} GAIA questions"
+@tool
+def search_wikipedia(query: str) -> str:
+    """Search Wikipedia for factual information."""
+    try:
+        # Wikipedia API search
+        search_url = "https://en.wikipedia.org/api/rest_v1/page/summary/" + query.replace(" ", "_")
+        response = requests.get(search_url, timeout=10)
+        
+        if response.status_code == 200:
+            data = response.json()
+            extract = data.get("extract", "")
+            if extract:
+                return f"Wikipedia: {extract[:500]}..."
             else:
-                return f"⚠️ Failed to fetch questions: {resp.status_code}"
-        except Exception as e:
-            return f"⚠️ Error fetching questions: {e}"
+                return f"Wikipedia page found for '{query}' but no extract available."
+        else:
+            return f"Wikipedia search failed for '{query}'"
+            
+    except Exception as e:
+        return f"Wikipedia search error: {str(e)}"
 
-    def fetch_random_question(self):
+
+@tool
+def execute_python(code: str) -> str:
+    """Execute Python code and return the result."""
+    try:
+        # Create a safe execution environment
+        safe_globals = {
+            '__builtins__': {
+                'print': print,
+                'len': len,
+                'str': str,
+                'int': int,
+                'float': float,
+                'bool': bool,
+                'list': list,
+                'dict': dict,
+                'tuple': tuple,
+                'set': set,
+                'range': range,
+                'sum': sum,
+                'max': max,
+                'min': min,
+                'abs': abs,
+                'round': round,
+                'sorted': sorted,
+                'enumerate': enumerate,
+                'zip': zip,
+            },
+            'math': __import__('math'),
+            'json': __import__('json'),
+            'datetime': __import__('datetime'),
+            'random': __import__('random'),
+        }
+        
+        # Capture output
+        import io
+        import sys
+        
+        old_stdout = sys.stdout
+        sys.stdout = mystdout = io.StringIO()
+        
         try:
-            resp = requests.get(f"{GAIA_API_BASE}/random-question", timeout=10)
-            if resp.status_code == 200:
-                return resp.json()
-            else:
-                return {}
-        except:
-            return {}
+            # Execute the code
+            exec(code, safe_globals)
+            output = mystdout.getvalue()
+        finally:
+            sys.stdout = old_stdout
+        
+        return output if output else "Code executed successfully (no output)"
+        
+    except Exception as e:
+        return f"Python execution error: {str(e)}"
 
-    def clean_answer(self, answer: str):
-        answer = answer.strip()
-        prefixes = ["Answer:", "Final answer:", "The answer is:"]
-        for prefix in prefixes:
-            if answer.lower().startswith(prefix.lower()):
-                answer = answer[len(prefix):].strip()
-        return answer.strip()
 
-    def answer_question(self, question_obj):
-        q = question_obj.get("Question", "")
-        system_prompt = (
-            "You are solving GAIA benchmark questions. "
-            "Provide ONLY the final answer, no reasoning."
+@tool
+def read_excel_file(file_path: str, sheet_name: Optional[str] = None) -> str:
+    """Read an Excel file and return its contents as a formatted string."""
+    try:
+        file_path_obj = Path(file_path)
+        if not file_path_obj.exists():
+            return f"Error: File not found at {file_path}"
+        
+        # Try to read the Excel file
+        if sheet_name and sheet_name.isdigit():
+            sheet_name = int(sheet_name)
+        elif sheet_name is None:
+            sheet_name = 0
+            
+        df = pd.read_excel(file_path, sheet_name=sheet_name)
+        
+        # Convert to string representation
+        if len(df) > 20:
+            # Show first 10 and last 10 rows for large datasets
+            result = f"Excel file with {len(df)} rows and {len(df.columns)} columns:\n\n"
+            result += "First 10 rows:\n"
+            result += df.head(10).to_string(index=False)
+            result += f"\n\n... ({len(df) - 20} rows omitted) ...\n\n"
+            result += "Last 10 rows:\n"
+            result += df.tail(10).to_string(index=False)
+        else:
+            result = f"Excel file with {len(df)} rows and {len(df.columns)} columns:\n\n"
+            result += df.to_string(index=False)
+            
+        return result
+        
+    except Exception as e:
+        return f"Error reading Excel file: {str(e)}"
+
+
+@tool
+def read_text_file(file_path: str) -> str:
+    """Read a text file and return its contents."""
+    try:
+        file_path_obj = Path(file_path)
+        if not file_path_obj.exists():
+            return f"Error: File not found at {file_path}"
+        
+        # Try different encodings
+        encodings = ['utf-8', 'utf-16', 'iso-8859-1', 'cp1252']
+        
+        for encoding in encodings:
+            try:
+                with open(file_path_obj, 'r', encoding=encoding) as f:
+                    content = f.read()
+                return f"File content ({encoding} encoding):\n\n{content}"
+            except UnicodeDecodeError:
+                continue
+        
+        return f"Error: Could not decode file with any standard encoding"
+        
+    except Exception as e:
+        return f"Error reading file: {str(e)}"
+
+
+class GaiaAgent:
+    """LangGraph-based agent for GAIA tasks using OpenRouter DeepSeek"""
+    
+    def __init__(self):
+        print("Initializing GaiaAgent with LangGraph and OpenRouter DeepSeek...")
+        
+        # Initialize the LLM
+        self.llm = OpenRouterLLM(
+            model="deepseek/deepseek-v3.1-terminus",
+            temperature=0.1,
+            max_tokens=2000
+        )
+        
+        # Define available tools
+        self.tools = [
+            search_web,
+            search_wikipedia,
+            execute_python,
+            read_excel_file,
+            read_text_file,
+        ]
+        
+        # Create the agent
+        self.agent = create_react_agent(
+            self.llm,
+            self.tools,
+            state_modifier=self._get_system_prompt()
         )
-        raw_answer = self.llm.generate(q, system_prompt)
-        return self.clean_answer(raw_answer)
-
-    def evaluate_all(self):
-        if not self.questions:
-            return {"error": "No questions loaded"}
-        results = []
-        correct = 0
-        for q in self.questions:
-            expected = str(q.get("Final answer", "")).strip()
-            answer = self.answer_question(q)
-            is_correct = answer.strip() == expected
-            if is_correct:
-                correct += 1
-            results.append({
-                "task_id": q.get("task_id"),
-                "question": q.get("Question"),
-                "expected": expected,
-                "answer": answer,
-                "correct": is_correct
-            })
-        score = (correct / len(results)) * 100 if results else 0
-        return {"score": score, "results": results, "correct": correct, "total": len(results)}
-
-    def submit_answers(self, username, agent_code, answers):
+        
+        print("GaiaAgent initialized successfully!")
+    
+    def _get_system_prompt(self) -> str:
+        """Get the system prompt for the agent"""
+        return """You are an advanced AI agent designed to answer complex questions that may require:
+
+1. Web searches for current information
+2. Mathematical calculations using Python
+3. File analysis (Excel, text files)  
+4. Multi-step reasoning and problem solving
+
+For GAIA evaluation:
+- Provide EXACT, DIRECT answers
+- Use tools when necessary to gather information or perform calculations
+- For math problems, show your calculation but end with just the number
+- For yes/no questions, answer just "Yes" or "No"
+- For factual questions, provide just the fact
+
+When you encounter files:
+- Use read_excel_file for .xlsx, .xls files
+- Use read_text_file for text-based files
+- Analyze the file content to answer the question
+
+Be thorough in your analysis but concise in your final answer."""
+
+    def __call__(self, task_id: str, question: str) -> str:
+        """Process a question and return the answer"""
         try:
-            payload = {
-                "username": username,
-                "agent_code": agent_code,
-                "answers": answers
-            }
-            resp = requests.post(f"{GAIA_API_BASE}/submit", json=payload, timeout=60)
-            if resp.status_code == 200:
-                return resp.json()
-            else:
-                return {"error": f"Submission failed: {resp.status_code}"}
+            print(f"Processing task {task_id}: {question[:100]}...")
+            
+            # Create the input state
+            messages = [HumanMessage(content=question)]
+            
+            # Run the agent
+            result = self.agent.invoke({"messages": messages})
+            
+            # Extract the final answer
+            final_message = result["messages"][-1]
+            answer = final_message.content
+            
+            # Clean up the answer for GAIA evaluation
+            clean_answer = self._clean_answer(answer)
+            
+            print(f"Agent answer for {task_id}: {clean_answer}")
+            return clean_answer
+            
         except Exception as e:
-            return {"error": str(e)}
-
-# ===============================
-# 3. Gradio UI
-# ===============================
-llm = OpenRouterLLM()
-agent = GAIAAgent(llm)
-
-def load_questions_ui():
-    return agent.fetch_questions()
-
-def test_random_question_ui():
-    q = agent.fetch_random_question()
-    if not q:
-        return "Failed to fetch a random question"
-    ans = agent.answer_question(q)
-    return f"Question: {q.get('Question')}\nAnswer: {ans}"
-
-def run_full_evaluation_ui(username):
-    if not agent.questions:
-        return "Please load questions first."
-    results_data = agent.evaluate_all()
-    if "error" in results_data:
-        return results_data["error"]
-
-    answers_payload = [
-        {"task_id": r["task_id"], "submitted_answer": r["answer"]}
-        for r in results_data["results"]
-    ]
-    agent_code = f"https://huggingface.co/spaces/{username}/Gaia-Test-Agent/tree/main"
-    submission_result = agent.submit_answers(username, agent_code, answers_payload)
-    score = submission_result.get("score", 0)
-    return f"Score: {score}%\nAnswers submitted: {len(answers_payload)}\nLeaderboard info: {submission_result}"
-
-def manual_test_ui(question_text):
-    return agent.answer_question({"Question": question_text})
-
-def build_gradio_app():
-    with gr.Blocks() as app:
-        gr.Markdown("# 🤖 GAIA Benchmark Agent")
-        
-        with gr.Tab("Load Questions"):
-            out_load = gr.Textbox(label="Status")
-            btn_load = gr.Button("Load GAIA Questions")
-            btn_load.click(load_questions_ui, outputs=out_load)
-        
-        with gr.Tab("Random Question Test"):
-            out_test = gr.Textbox(label="Result", lines=6)
-            btn_test = gr.Button("Test Random Question")
-            btn_test.click(test_random_question_ui, outputs=out_test)
-        
-        with gr.Tab("Full Evaluation & Submit"):
-            username_input = gr.Textbox(label="Your HF Username")
-            out_eval = gr.Textbox(label="Evaluation Result", lines=10)
-            btn_eval = gr.Button("Run Evaluation & Submit")
-            btn_eval.click(run_full_evaluation_ui, inputs=username_input, outputs=out_eval)
-        
-        with gr.Tab("Manual Test"):
-            manual_input = gr.Textbox(label="Enter Question")
-            manual_output = gr.Textbox(label="Agent Answer", lines=4)
-            manual_btn = gr.Button("Get Answer")
-            manual_btn.click(manual_test_ui, inputs=manual_input, outputs=manual_output)
-    return app
-
-# ===============================
-# 4. Main
-# ===============================
-if __name__ == "__main__":
-    app = build_gradio_app()
-    if os.getenv("SPACE_ID"):
-        app.launch(server_name="0.0.0.0", server_port=7860)
-    else:
-        app.launch(share=True)
+            error_msg = f"Agent error: {str(e)}"
+            print(f"Error processing task {task_id}: {error_msg}")
+            return error_msg
+    
+    def _clean_answer(self, answer: str) -> str:
+        """Clean the answer to extract the final result"""
+        answer = answer.strip()
+        
+        # Look for "Final Answer:" pattern
+        if "final answer:" in answer.lower():
+            parts = re.split(r'final answer:', answer, flags=re.IGNORECASE)
+            if len(parts) > 1:
+                answer = parts[-1].strip()
+        
+        # Remove common prefixes
+        prefixes = [
+            "The answer is", "Answer:", "Result:", "Solution:",
+            "Based on", "Therefore", "In conclusion", "So the answer is"
+        ]
+        
+        for prefix in prefixes:
+            if answer.lower().startswith(prefix.lower()):
+                answer = answer[len(prefix):].strip()
+                if answer.startswith(':'):
+                    answer = answer[1:].strip()
+                break
+        
+        # Remove quotes and periods from short answers
+        if len(answer.split()) <= 3:
+            answer = answer.strip('"\'.')
+        
+        return answer