Update app.py

app.py

@@ -9,25 +9,26 @@ from audio_transcriber import AudioTranscriptionTool
 from image_analyzer import ImageAnalysisTool
 from wikipedia_searcher import WikipediaSearcher
 
-# GAIA scoring endpoint
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-GAIA_SYSTEM_PROMPT = """You are an agent solving the GAIA benchmark and you are required to provide exact answers.
-Rules to follow:
-1. Return only the exact requested answer: no explanation and no reasoning.
-2. For yes/no questions, return exactly \"Yes\" or \"No\".
-3. For dates, use the exact format requested.
-4. For numbers, use the exact number, no other format.
-5. For names, use the exact name as found in sources.
-6. If the question has an associated file, download the file first using the task ID.
-Examples of good responses:
-- \"42\"
-- \"Arturo Nunez\"
-- \"Yes\"
-- \"October 5, 2001\"
-- \"Buenos Aires\"
-Never include phrases like \"the answer is...\" or \"Based on my research\".
-Only return the exact answer."""
+# Zephyr-compatible system prompt to prepend manually
+SYSTEM_PROMPT = (
+    "You are an agent solving the GAIA benchmark and must provide exact answers.\n"
+    "Rules:\n"
+    "1. Return only the exact requested answer: no explanation.\n"
+    "2. For yes/no, return 'Yes' or 'No'.\n"
+    "3. For dates, use the exact requested format.\n"
+    "4. For numbers, use only the number.\n"
+    "5. For names, use the exact name from sources.\n"
+    "6. If the question has a file, download it using the task ID.\n"
+    "Examples:\n"
+    "- '42'\n"
+    "- 'Arturo Nunez'\n"
+    "- 'Yes'\n"
+    "- 'October 5, 2001'\n"
+    "- 'Buenos Aires'\n"
+    "Never say 'the answer is...'. Only return the answer.\n"
+)
 
 class GaiaAgent:
     def __init__(self):
@@ -52,11 +53,10 @@ class GaiaAgent:
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
 
+        full_prompt = f"{SYSTEM_PROMPT}\nQUESTION:\n{question}"
+
         try:
-            result = self.agent.run(
-                question,
-                system_prompt=GAIA_SYSTEM_PROMPT
-            )
+            result = self.agent.run(full_prompt)
             print(f"Raw result from agent: {result}")
 
             if isinstance(result, dict) and "answer" in result:
@@ -97,7 +97,6 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
     print(f"Agent code URL: {agent_code}")
 
-    print(f"Fetching questions from: {questions_url}")
     try:
         response = requests.get(questions_url, timeout=15)
         response.raise_for_status()
@@ -110,15 +109,13 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
 
     results_log = []
     answers_payload = []
-    print(f"Running agent on {len(questions_data)} questions...")
+
     for item in questions_data:
         task_id = item.get("task_id")
         if not task_id:
             continue
         try:
             submitted_answer = agent(item.get("question", ""))
-            print(f"Q: {item.get('question', '')[:60]}...")
-            print(f"A: {submitted_answer}\n")
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({
                 "Task ID": task_id,