Update app.py

app.py

@@ -9,7 +9,7 @@ from duckduckgo_search import DDGS
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
-DELAY_BETWEEN_QUESTIONS = 8  # 8 seconds between questions to avoid rate limits
+DELAY_BETWEEN_QUESTIONS = 15  # 15 seconds to avoid rate limits on 70B model
 
 # ============================================
 # GROQ CLIENT
@@ -28,12 +28,17 @@ def get_groq_client():
 def web_search(query: str) -> str:
     try:
         with DDGS() as ddgs:
-            results = list(ddgs.text(query, max_results=3))
+            results = list(ddgs.text(query, max_results=5))
         if not results:
-            return "No results"
-        return "\n".join([f"- {r.get('title','')}: {r.get('body','')}" for r in results])
+            return ""
+        output = []
+        for r in results:
+            output.append(f"Title: {r.get('title','')}")
+            output.append(f"Snippet: {r.get('body','')}")
+            output.append("---")
+        return "\n".join(output)
     except:
-        return "Search failed"
+        return ""
 
 
 def get_task_file(task_id: str) -> dict:
@@ -51,26 +56,41 @@ def get_task_file(task_id: str) -> dict:
         if 'filename=' in disposition:
             filename = disposition.split('filename=')[-1].strip('"\'')
         
+        result = {"has_file": True, "filename": filename, "type": content_type}
+        
+        # Python files - return code
+        if filename.endswith('.py'):
+            result["content"] = response.text
+            result["is_code"] = True
+            return result
+        
         # Text files
-        if 'text' in content_type or filename.endswith(('.txt', '.py', '.md', '.csv', '.json')):
-            return {"has_file": True, "content": response.text[:5000]}
+        if 'text' in content_type or filename.endswith(('.txt', '.md', '.csv', '.json')):
+            result["content"] = response.text[:6000]
+            return result
         
         # Excel
         if 'excel' in content_type or 'spreadsheet' in content_type or filename.endswith(('.xlsx', '.xls')):
             try:
                 from io import BytesIO
                 df = pd.read_excel(BytesIO(response.content))
-                return {"has_file": True, "content": df.to_string()}
-            except:
-                return {"has_file": True, "content": "[Excel file - parse failed]"}
+                result["content"] = df.to_string()
+                result["is_excel"] = True
+                return result
+            except Exception as e:
+                result["content"] = f"[Excel parse error: {e}]"
+                return result
         
-        # Image
+        # Image - can't process
         if 'image' in content_type:
-            return {"has_file": True, "content": "[IMAGE - cannot process]", "is_image": True}
+            result["content"] = "[IMAGE FILE]"
+            result["is_image"] = True
+            return result
         
-        return {"has_file": True, "content": f"[Binary: {content_type}]"}
-    except:
-        return {"has_file": False}
+        result["content"] = f"[File: {content_type}]"
+        return result
+    except Exception as e:
+        return {"has_file": False, "error": str(e)}
 
 
 def reverse_string(text: str) -> str:
@@ -78,7 +98,26 @@ def reverse_string(text: str) -> str:
 
 
 def is_reversed(text: str) -> bool:
-    return any(x in text.lower() for x in ['.rewsna', 'eht sa', 'tfel', 'drow eht'])
+    indicators = ['.rewsna', 'eht sa', 'tfel', 'drow eht', 'etisoppo']
+    return any(x in text.lower() for x in indicators)
+
+
+def execute_python(code: str) -> str:
+    """Safely execute Python code and return output"""
+    try:
+        import io
+        import sys
+        from contextlib import redirect_stdout
+        
+        # Capture stdout
+        f = io.StringIO()
+        with redirect_stdout(f):
+            exec(code, {"__builtins__": __builtins__})
+        
+        output = f.getvalue()
+        return output.strip() if output else "No output"
+    except Exception as e:
+        return f"Error: {e}"
 
 
 # ============================================
@@ -87,61 +126,119 @@ def is_reversed(text: str) -> bool:
 
 class BasicAgent:
     def __init__(self):
-        print("Initializing Groq agent...")
+        print("Initializing Groq agent (70B model)...")
         self.client = get_groq_client()
         print("✅ Ready!")
     
-    def ask(self, prompt: str) -> str:
-        for attempt in range(3):
+    def ask(self, prompt: str, max_retries: int = 3) -> str:
+        for attempt in range(max_retries):
             try:
                 response = self.client.chat.completions.create(
-                    model="llama-3.1-8b-instant",  # Faster model with better rate limits
+                    model="llama-3.3-70b-versatile",  # Smart model
                     messages=[{"role": "user", "content": prompt}],
                     temperature=0,
-                    max_tokens=100,
+                    max_tokens=150,
                 )
                 return response.choices[0].message.content.strip()
             except Exception as e:
                 if "rate" in str(e).lower() or "429" in str(e):
-                    wait = (attempt + 1) * 15
+                    wait = (attempt + 1) * 20
                     print(f"    ⏳ Rate limit, waiting {wait}s...")
                     time.sleep(wait)
                 else:
                     return f"Error: {e}"
         return "unknown"
     
+    def clean_answer(self, answer: str) -> str:
+        # Remove common prefixes
+        prefixes = [
+            "Answer:", "The answer is:", "The answer is", "A:", 
+            "Final answer:", "Final answer", "Based on",
+            "I found that", "The result is", "**", "```"
+        ]
+        for p in prefixes:
+            if answer.lower().startswith(p.lower()):
+                answer = answer[len(p):].strip()
+        
+        # Remove markdown and quotes
+        answer = answer.replace("**", "").replace("```", "").strip()
+        answer = answer.strip('"\'')
+        
+        # If answer is too long or contains "I'm unable", return unknown
+        if "I'm unable" in answer or "I cannot" in answer or "I don't" in answer:
+            return "unknown"
+        
+        # Remove trailing period for short answers
+        if answer.endswith('.') and len(answer.split()) <= 5:
+            answer = answer[:-1]
+        
+        return answer.strip()
+    
     def __call__(self, question: str, task_id: str = None) -> str:
-        context = ""
+        original_question = question
+        context_parts = []
         
-        # Handle reversed text
+        # 1. Handle reversed text
         if is_reversed(question):
             question = reverse_string(question)
+            context_parts.append(f"[Original was reversed. Decoded: {question}]")
         
-        # Get file if exists
+        # 2. Check for file
+        file_info = {"has_file": False}
         if task_id:
-            file = get_task_file(task_id)
-            if file.get("has_file") and file.get("content"):
-                context = f"FILE: {file['content']}\n\n"
+            file_info = get_task_file(task_id)
+            
+            if file_info.get("has_file"):
+                if file_info.get("is_code"):
+                    # Execute Python code
+                    code = file_info.get("content", "")
+                    output = execute_python(code)
+                    context_parts.append(f"Python code output: {output}")
+                    
+                elif file_info.get("is_excel"):
+                    context_parts.append(f"Excel data:\n{file_info.get('content', '')[:3000]}")
+                    
+                elif file_info.get("is_image"):
+                    context_parts.append("[This task has an image file which cannot be processed]")
+                    
+                else:
+                    context_parts.append(f"File content:\n{file_info.get('content', '')[:3000]}")
         
-        # Search if needed and no file
-        if not context:
-            keywords = ["who", "what", "when", "where", "how many", "album", "actor", "surname", "wikipedia"]
-            if any(k in question.lower() for k in keywords):
-                results = web_search(question[:80])
-                if results != "Search failed":
-                    context = f"INFO: {results}\n\n"
+        # 3. Web search if needed (and no useful file)
+        if not file_info.get("has_file") or file_info.get("is_image"):
+            search_triggers = [
+                "who ", "what ", "when ", "where ", "how many", "how much",
+                "album", "actor", "movie", "wikipedia", "surname", "name",
+                "athlete", "pitcher", "yankee", "country", "competition",
+                "nominated", "published", "article", "mercedes", "sosa"
+            ]
+            
+            if any(t in question.lower() for t in search_triggers):
+                search_results = web_search(question)
+                if search_results:
+                    context_parts.append(f"Search results:\n{search_results[:2500]}")
         
-        prompt = f"{context}Q: {question}\n\nAnswer with ONLY the final answer, nothing else:"
+        # 4. Build prompt
+        context = "\n\n".join(context_parts) if context_parts else ""
         
+        prompt = f"""You must answer this question with ONLY the final answer.
+
+RULES:
+- Give ONLY the answer (a word, number, name, or short phrase)
+- NO explanations, NO "I think", NO "Based on"
+- If asked for a number, give just the number
+- If asked for a name, give just the name
+- If it's a list, give comma-separated items
+- NEVER say "I'm unable to" or "I cannot" - give your best guess
+
+{f"CONTEXT:{chr(10)}{context}" if context else ""}
+
+QUESTION: {question}
+
+YOUR ANSWER (just the answer):"""
+
         answer = self.ask(prompt)
-        
-        # Clean
-        for p in ["Answer:", "The answer is:", "A:", "Final answer:"]:
-            if answer.lower().startswith(p.lower()):
-                answer = answer[len(p):].strip()
-        answer = answer.strip('"\'.')
-        
-        return answer
+        return self.clean_answer(answer)
 
 
 # ============================================
@@ -167,7 +264,8 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     try:
         questions = requests.get(f"{DEFAULT_API_URL}/questions", timeout=15).json()
-        print(f"📋 {len(questions)} questions\n")
+        print(f"📋 {len(questions)} questions")
+        print(f"⏱️ Expected time: ~{len(questions) * DELAY_BETWEEN_QUESTIONS // 60} minutes\n")
     except Exception as e:
         return f"❌ Fetch failed: {e}", None
 
@@ -182,16 +280,18 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         print(f"[{i+1}/{len(questions)}] {question[:50]}...")
         
         answer = agent(question, task_id)
-        print(f"    → {answer[:40]}")
+        print(f"    → {answer[:50]}")
         
         answers.append({"task_id": task_id, "submitted_answer": answer})
         results.append({"#": i+1, "Q": question[:40]+"...", "A": answer[:50]})
         
+        # Delay to avoid rate limits
         if i < len(questions) - 1:
+            print(f"    ⏳ Waiting {DELAY_BETWEEN_QUESTIONS}s...")
             time.sleep(DELAY_BETWEEN_QUESTIONS)
 
     total = time.time() - start
-    print(f"\n⏱️ {total:.0f}s")
+    print(f"\n⏱️ Total: {total:.0f}s ({total/60:.1f} min)")
 
     try:
         result = requests.post(
@@ -208,7 +308,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         correct = result.get('correct_count', 0)
         total_q = result.get('total_attempted', 0)
         
-        status = f"✅ Done in {total:.0f}s\n\n🎯 {score}% ({correct}/{total_q})\n\n"
+        status = f"✅ Done in {total/60:.1f} min\n\n🎯 {score}% ({correct}/{total_q})\n\n"
         status += "🎉 PASSED!" if score >= 30 else f"Need {30-score}% more"
         
         return status, pd.DataFrame(results)
@@ -221,8 +321,12 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 # ============================================
 
 with gr.Blocks() as demo:
-    gr.Markdown("# 🎯 GAIA Agent")
-    gr.Markdown("Groq + Llama 3.1 8B (fast, good rate limits)")
+    gr.Markdown("# 🎯 GAIA Agent - Unit 4")
+    gr.Markdown("""
+    **Groq + Llama 3.3 70B** (smart model)
+    
+    ⏱️ Takes ~5 minutes (15s delay between questions to avoid rate limits)
+    """)
     
     gr.LoginButton()
     run_btn = gr.Button("🚀 Run", variant="primary", size="lg")