Update app.py

app.py

@@ -9,16 +9,32 @@ 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"
 
-# Define the GaiaAgent class with embedded prompt in __call__
+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.
+"""
+
 class GaiaAgent:
     def __init__(self):
         print("Gaia Agent Initialized")
 
         self.model = InferenceClientModel(
-            model_id="cognitivecomputations/dolphin-2.6-mixtral-8x7b",
+            model_id="mistralai/Mistral-7B-Instruct-v0.2",
             token=os.getenv("HF_API_TOKEN", "").strip()
         )
 
@@ -33,31 +49,29 @@ class GaiaAgent:
             model=self.model
         )
 
-    def __call__(self, question: str) -> str:
+    def __call__(self, question: str, task_id: str = "") -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
 
-        prompt = f"""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.
-QUESTION:
-{question}
-"""
+        file_path = None
+        if task_id:
+            try:
+                file_url = f"https://agents-course-unit4-scoring.hf.space/file={task_id}"
+                print(f"Attempting to download file from {file_url}")
+                response = requests.get(file_url)
+                response.raise_for_status()
+                file_path = f"/tmp/{task_id}"
+                with open(file_path, "wb") as f:
+                    f.write(response.content)
+                print(f"Downloaded file for task {task_id} to {file_path}")
+            except Exception as e:
+                print(f"Warning: Failed to download file for {task_id}: {e}")
 
         try:
-            result = self.agent.run(prompt)
+            result = self.agent.run(
+                input=question,
+                system_prompt=GAIA_SYSTEM_PROMPT,
+                files=[file_path] if file_path else None
+            )
             print(f"Raw result from agent: {result}")
 
             if isinstance(result, dict) and "answer" in result:
@@ -75,7 +89,6 @@ QUESTION:
             print(f"Exception during agent run: {e}")
             return f"AGENT ERROR: {e}"
 
-# Evaluation + Submission function
 def run_and_submit_all(profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID")
 
@@ -114,22 +127,23 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     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:
+        task_id = item.get("task_id", "")
+        question = item.get("question", "")
+        if not question:
             continue
         try:
-            submitted_answer = agent(item.get("question", ""))
+            submitted_answer = agent(question, task_id)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({
                 "Task ID": task_id,
-                "Question": item.get("question", ""),
+                "Question": question,
                 "Submitted Answer": submitted_answer
             })
         except Exception as e:
             error_msg = f"AGENT ERROR: {e}"
             results_log.append({
                 "Task ID": task_id,
-                "Question": item.get("question", ""),
+                "Question": question,
                 "Submitted Answer": error_msg
             })
 
@@ -167,15 +181,14 @@ def run_and_submit_all(profile: gr.OAuthProfile | None):
     except Exception as e:
         return f"An unexpected error occurred during submission: {e}", pd.DataFrame(results_log)
 
-# Gradio UI
 with gr.Blocks() as demo:
     gr.Markdown("# Basic Agent Evaluation Runner")
-    gr.Markdown("""
-        **Instructions:**
-        1. Clone this space and define your agent and tools.
-        2. Log in to your Hugging Face account using the button below.
-        3. Click 'Run Evaluation & Submit All Answers' to test your agent and submit results.
-    """)
+    gr.Markdown("""\
+**Instructions:**
+1. Clone this space and define your agent and tools.
+2. Log in to your Hugging Face account using the button below.
+3. Click 'Run Evaluation & Submit All Answers' to test your agent and submit results.
+""")
 
     gr.LoginButton()
     run_button = gr.Button("Run Evaluation & Submit All Answers")