Update app.py

app.py

@@ -4,24 +4,29 @@ import requests
 import pandas as pd
 import tempfile
 import traceback
+import subprocess
+import sys
 
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # --- File helpers ---
 def download_task_file(task_id: str):
+    """Returns (local_path, filename) or (None, None)."""
     url = f"{DEFAULT_API_URL}/files/{task_id}"
     try:
         resp = requests.get(url, timeout=30)
         if resp.status_code != 200:
+            print(f"No file for task {task_id}: HTTP {resp.status_code}")
             return None, None
         cd = resp.headers.get("content-disposition", "")
         fname = "task_file"
         if "filename=" in cd:
-            fname = cd.split("filename=")[-1].strip().strip('"')
+            fname = cd.split("filename=")[-1].strip().strip('"').strip("'")
         ext = os.path.splitext(fname)[-1] or ".bin"
         tmp = tempfile.NamedTemporaryFile(delete=False, suffix=ext)
         tmp.write(resp.content)
         tmp.close()
+        print(f"Downloaded file: {fname} -> {tmp.name} ({len(resp.content)} bytes)")
         return tmp.name, fname
     except Exception as e:
         print(f"File download error: {e}")
@@ -39,11 +44,32 @@ def read_file_contents(local_path: str, fname: str) -> str:
         elif ext in (".py", ".txt", ".md", ".json"):
             with open(local_path) as f:
                 return f.read()
+        elif ext in (".png", ".jpg", ".jpeg", ".gif", ".webp"):
+            return f"[IMAGE FILE: {fname}] - This is an image that needs visual analysis."
         else:
-            return f"Binary file: {fname}"
+            # Try reading as text anyway
+            try:
+                with open(local_path) as f:
+                    return f.read()
+            except:
+                return f"Binary file: {fname} ({ext})"
     except Exception as e:
         return f"Error reading file: {e}"
 
+def run_python_file(local_path: str) -> str:
+    try:
+        result = subprocess.run(
+            [sys.executable, local_path],
+            capture_output=True, text=True, timeout=15
+        )
+        output = result.stdout + result.stderr
+        print(f"Python output: {output[:300]}")
+        return output.strip() if output.strip() else "No output produced."
+    except subprocess.TimeoutExpired:
+        return "Code execution timed out."
+    except Exception as e:
+        return f"Execution error: {e}"
+
 def clean_answer(text: str) -> str:
     text = text.strip()
     for prefix in ["FINAL ANSWER:", "Final Answer:", "Answer:",
@@ -53,69 +79,59 @@ def clean_answer(text: str) -> str:
             text = text[len(prefix):].strip()
     return text.split("\n")[0].strip().strip('"').strip("'").strip("*").strip()
 
-# --- Groq API call (free tier, very fast) ---
-def call_groq(api_key: str, prompt: str, system: str = "") -> str:
+# --- Groq API ---
+def call_groq(api_key: str, prompt: str, system: str = "", max_tokens: int = 512) -> str:
     url = "https://api.groq.com/openai/v1/chat/completions"
-    headers = {
-        "Authorization": f"Bearer {api_key}",
-        "Content-Type": "application/json"
-    }
+    headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
     messages = []
     if system:
         messages.append({"role": "system", "content": system})
     messages.append({"role": "user", "content": prompt})
-
-    body = {
-        "model": "llama-3.3-70b-versatile",
-        "messages": messages,
-        "temperature": 0.0,
-        "max_tokens": 512,
-    }
+    body = {"model": "llama-3.3-70b-versatile", "messages": messages,
+            "temperature": 0.0, "max_tokens": max_tokens}
     resp = requests.post(url, headers=headers, json=body, timeout=60)
     print(f"Groq status: {resp.status_code}")
     if resp.status_code != 200:
         print(f"Groq error: {resp.text[:400]}")
         raise Exception(f"Groq API error {resp.status_code}: {resp.text[:200]}")
-    data = resp.json()
-    return data["choices"][0]["message"]["content"].strip()
-
-# --- Test function ---
-def test_api():
-    api_key = os.getenv("GROQ_API_KEY", "")
-    if not api_key:
-        return "❌ GROQ_API_KEY is NOT set in Space Secrets!\nGet a free key at https://console.groq.com"
-    try:
-        answer = call_groq(api_key, "What is 2+2? Reply with just the number.", 
-                          "Reply with only the bare answer.")
-        return f"✅ Groq API working! Test answer: '{answer}'"
-    except Exception as e:
-        return f"❌ Groq failed: {e}"
+    return resp.json()["choices"][0]["message"]["content"].strip()
 
 # --- Web search ---
-def search_web(query: str) -> str:
+def search_web(query: str, max_results: int = 6) -> str:
     try:
         from duckduckgo_search import DDGS
         with DDGS() as ddgs:
-            results = list(ddgs.text(query, max_results=5))
+            results = list(ddgs.text(query, max_results=max_results))
         if not results:
             return "No results found."
         return "\n\n".join(
-            f"Title: {r.get('title','')}\nSnippet: {r.get('body','')}"
+            f"Title: {r.get('title','')}\nSnippet: {r.get('body','')}\nURL: {r.get('href','')}"
             for r in results
         )
     except Exception as e:
         return f"Search error: {e}"
 
-# --- Agent ---
-SYSTEM_PROMPT = """You are an expert AI agent solving GAIA benchmark questions.
+def test_api():
+    api_key = os.getenv("GROQ_API_KEY", "")
+    if not api_key:
+        return "❌ GROQ_API_KEY is NOT set in Space Secrets!"
+    try:
+        answer = call_groq(api_key, "What is 2+2? Reply with just the number.", "Reply with only the bare answer.")
+        return f"✅ Groq API working! Test answer: '{answer}'"
+    except Exception as e:
+        return f"❌ Groq failed: {e}"
 
-RULES:
-- Reply with ONLY the final answer. No explanation, no preamble.
-- No "The answer is", no "FINAL ANSWER", no prefixes at all.
-- Give the shortest correct answer: a name, number, word, or short phrase.
-- Exact match grading is used — precision matters.
-- For numbers: use digits unless words are specifically asked for.
-- For lists: comma-separated values.
+# --- System prompt ---
+SYSTEM_PROMPT = """You are an expert AI agent solving GAIA benchmark questions. Exact match grading is used.
+
+CRITICAL RULES:
+1. Reply with ONLY the final answer — no explanation, no preamble, no prefix like "The answer is"
+2. Be as concise as possible: just the name, number, word, or short phrase
+3. For numbers: use digits (e.g. "42") unless words are specifically requested
+4. For currency: strip $ signs and commas unless format is specifically asked for (e.g. "1234.56" not "$1,234.56")
+5. For lists: use comma-separated values with no extra words
+6. For names: give full name in the exact format requested (first name only if asked for first name)
+7. Think carefully — precision matters for exact matching
 """
 
 class BasicAgent:
@@ -126,8 +142,7 @@ class BasicAgent:
                 "GROQ_API_KEY not set!\n"
                 "1. Go to https://console.groq.com → free account → API Keys → Create key\n"
                 "2. Space Settings → Variables and Secrets → New Secret\n"
-                "   Name: GROQ_API_KEY  Value: your key\n"
-                "3. Restart Space"
+                "   Name: GROQ_API_KEY  Value: your key"
             )
         print(f"BasicAgent ready. Key: {self.api_key[:8]}...")
 
@@ -141,68 +156,87 @@ class BasicAgent:
 
         print(f"\n{'='*50}\nTask: {task_id}\nQ: {question[:200]}")
 
-        # 1. Download attached file
         file_context = ""
-        if task_id:
-            local_path, fname = download_task_file(task_id)
-            if local_path:
-                contents = read_file_contents(local_path, fname)
-                file_context = f"\n\n[Attached file: {fname}]\n{contents[:4000]}\n[End of file]\n"
-                print(f"Got file: {fname}")
-
-        # 2. Execute Python if it's a code file
         code_output = ""
-        if task_id and file_context and fname.endswith(".py"):
-            try:
-                import subprocess, sys
-                result = subprocess.run(
-                    [sys.executable, local_path],
-                    capture_output=True, text=True, timeout=15
-                )
-                code_output = f"\n\n[Python execution output]\n{result.stdout}\n{result.stderr}\n[End output]\n"
-                print(f"Code output: {result.stdout[:200]}")
-            except Exception as e:
-                code_output = f"\n[Code execution error: {e}]\n"
+        local_path = None
+        fname = None
 
-        # 3. Web search for factual questions
+        # 1. Always try to download file for every task
+        if task_id:
+            local_path, fname = download_task_file(task_id)
+            if local_path and fname:
+                ext = os.path.splitext(fname)[-1].lower()
+
+                if ext == ".py":
+                    # Run Python code and capture output
+                    code_output_text = run_python_file(local_path)
+                    file_contents = read_file_contents(local_path, fname)
+                    file_context = (
+                        f"\n\n[Python file: {fname}]\n"
+                        f"CODE:\n{file_contents}\n\n"
+                        f"EXECUTION OUTPUT:\n{code_output_text}\n"
+                        f"[End of file]\n"
+                    )
+                elif ext in (".xlsx", ".xls", ".csv"):
+                    contents = read_file_contents(local_path, fname)
+                    file_context = f"\n\n[Data file: {fname}]\n{contents[:5000]}\n[End of file]\n"
+                elif ext in (".png", ".jpg", ".jpeg"):
+                    file_context = f"\n\n[Note: An image file '{fname}' is attached but cannot be displayed in text. Use your knowledge to answer based on the question context.]\n"
+                else:
+                    contents = read_file_contents(local_path, fname)
+                    file_context = f"\n\n[Attached file: {fname}]\n{contents[:4000]}\n[End of file]\n"
+
+        # 2. Web search — always search unless we have a code execution result
         search_context = ""
-        if not file_context:
-            print("Searching web...")
-            results = search_web(question[:200])
-            if results and "error" not in results.lower():
-                search_context = f"\n\n[Web search results]\n{results[:2500]}\n[End search]\n"
-
-        # 4. Build prompt and ask Groq
+        has_code_answer = local_path and fname and os.path.splitext(fname)[-1].lower() == ".py"
+
+        if not has_code_answer:
+            # Build a focused search query
+            search_query = question[:200]
+            print(f"Searching: {search_query[:80]}...")
+            results = search_web(search_query)
+            if results and "error" not in results.lower() and "No results" not in results:
+                search_context = f"\n\n[Web search results]\n{results[:3000]}\n[End search]\n"
+
+        # 3. Special handling for reversed text question
+        if "rewsna" in question or "dnatsrednu" in question:
+            # This is a reversed text question — reverse it first
+            reversed_q = question[::-1]
+            print(f"Reversed question: {reversed_q}")
+            question = reversed_q
+
+        # 4. Build prompt
         prompt = (
             f"Question: {question}"
             f"{file_context}"
-            f"{code_output}"
             f"{search_context}"
-            "\n\nProvide ONLY the final answer. Nothing else."
+            "\n\nProvide ONLY the final answer. No explanation. No prefix."
         )
 
         try:
-            answer = call_groq(self.api_key, prompt, SYSTEM_PROMPT)
+            answer = call_groq(self.api_key, prompt, SYSTEM_PROMPT, max_tokens=256)
             print(f"Raw answer: '{answer}'")
 
-            # If too verbose, ask to extract
-            if len(answer.split()) > 20:
+            # If too verbose, extract key part
+            if len(answer.split()) > 25:
                 answer = call_groq(
                     self.api_key,
-                    f"From this text, extract ONLY the shortest possible final answer "
-                    f"(name, number, or phrase). Nothing else:\n\n{answer}",
-                    "Reply with only the bare answer."
+                    f"From this response, extract ONLY the shortest final answer "
+                    f"(name, number, or brief phrase). Nothing else:\n\n{answer}",
+                    "Reply with only the bare answer. No explanation.",
+                    max_tokens=64
                 )
                 print(f"Extracted: '{answer}'")
 
             answer = clean_answer(answer)
             print(f"Final: '{answer}'")
             return answer
+
         except Exception as e:
             print(f"Agent error: {e}\n{traceback.format_exc()}")
             return ""
 
-# --- Submit function ---
+# --- Submit ---
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID")
     if not profile:
@@ -215,8 +249,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
         agent = BasicAgent()
     except RuntimeError as e:
         return f"❌ {e}", None
-    except Exception as e:
-        return f"Error initializing agent: {e}", None
 
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
 
@@ -276,13 +308,9 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 with gr.Blocks() as demo:
     gr.Markdown("# Basic Agent Evaluation Runner")
     gr.Markdown("""
-**One-time FREE setup:**
-1. Go to [console.groq.com](https://console.groq.com) → Sign up free → API Keys → **Create API Key**
-2. In your Space → **Settings → Variables and Secrets → New Secret**
-   - Name: `GROQ_API_KEY` | Value: paste your key
-3. Restart Space, then click **Test** below to confirm it works
+**Setup:** Add `GROQ_API_KEY` in Space Settings → Variables and Secrets → New Secret.
+Free key at [console.groq.com](https://console.groq.com)
 """)
-
     gr.LoginButton()
 
     with gr.Row():