update

app.py

@@ -23,64 +23,12 @@ from smolagents import (
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-
 @tool
-def get_image(file_name: str) -> str:
+def get_file(question_id: str, file_name: str) -> str:
     """
-    A tool that downloads an image from file name and saves it locally.
-    Args:
-        file_name: File name.
-    Returns:
-        str: Local file path where the image was saved.
-    """
-    import requests
-    import os
-    
-    url = f"{DEFAULT_API_URL}/files/{file_name}"
-    print(f"Fetching image from URL: {url}")
-    
-    # Create downloads directory if it doesn't exist
-    downloads_dir = "downloaded_images"
-    os.makedirs(downloads_dir, exist_ok=True)
-    
-    response = None
-    try:
-        response = requests.get(url, timeout=30)
-        response.raise_for_status()  # Raises an HTTPError for bad responses
-        
-        # Check if response is empty
-        if not response.content:
-            raise ValueError(f"Empty response received from {url}")
-        
-        # Check content type
-        content_type = response.headers.get('content-type', '').lower()
-        print(f"Response content-type: {content_type}")
-        print(f"Response content length: {len(response.content)} bytes")
-        
-        # Use original filename directly
-        local_path = os.path.join(downloads_dir, file_name)
-        
-        # Save the image to local file
-        with open(local_path, 'wb') as f:
-            f.write(response.content)
-        
-        print(f"Image saved to: {local_path}")
-        return local_path
-        
-    except requests.exceptions.RequestException as e:
-        raise ValueError(f"Failed to fetch image from {url}: {e}")
-    except Exception as e:
-        # Print first 200 characters of response content for debugging
-        content_preview = response.content[:200] if response and hasattr(response, 'content') else b"No response"
-        print(f"Error downloading image. Content preview: {content_preview}")
-        raise ValueError(f"Failed to download image from {url}: {e}")
-
-
-@tool
-def get_text_file(file_name: str) -> str:
-    """
-    A tool that downloads a text file (such as code) from file name and saves it locally.
+    A tool that downloads a file that was mentioned in a question.
     Args:
+        question_id: Question ID.
         file_name: File name.
     Returns:
         str: Local file path where the text was saved.
@@ -88,7 +36,7 @@ def get_text_file(file_name: str) -> str:
     import requests
     import os
     
-    url = f"{DEFAULT_API_URL}/files/{file_name}"
+    url = f"{DEFAULT_API_URL}/files/{question_id}"
     print(f"Fetching text file from URL: {url}")
     
     # Create downloads directory if it doesn't exist
@@ -134,7 +82,7 @@ class BasicAgent:
     def __init__(self):
         print("BasicAgent initialized.")
         self.multimodal_agent = CodeAgent(
-            tools=[VisitWebpageTool(), DuckDuckGoSearchTool()],
+            tools=[VisitWebpageTool(), DuckDuckGoSearchTool(), get_file],
             model= OpenAIServerModel(model_id="gpt-4o"),
             additional_authorized_imports=[
                 "requests",
@@ -151,19 +99,58 @@ class BasicAgent:
             ],
             name="multimodal_agent",
             description="""
-                I can understand natural language and I am optimised for visual recognition, image reasoning, captioning, and answering general questions about an image.""",
+                This agent can reason across audio, vision, and text""",
             verbosity_level=0,
             max_steps=10,
         ) 
         
-        self.manager_agent = CodeAgent(
-            tools=[VisitWebpageTool(), DuckDuckGoSearchTool(), get_image, get_text_file],
+        self.code_agent = CodeAgent(
+            tools=[VisitWebpageTool(), DuckDuckGoSearchTool(), get_file],
             model=InferenceClientModel(
                 model_id="Qwen/Qwen2.5-Coder-32B-Instruct",
             ),
-            managed_agents=[
-                self.multimodal_agent
+            additional_authorized_imports=[
+                "requests",
+                "bs4",
+                "markdownify",
+                "wikipedia",
+                "pandas",
+                "io",
+                "PIL",
+                "chess",
+                "img2text",
+                "chess.pgn",
+                "PIL.Image",
+                "bytes",
+                "cv2",
+                "numpy",
+                "chess.engine",
             ],
+            name="code_agent",
+            description="""
+                This agent specializes at: 
+                    - Writing code to solve problem. 
+                    - Browse the web to find information.
+                    - Solving chess problems.
+                This agent follow rules below when possible:
+                    1. `pandas` Python package is provided. Please try to use it FIRST when there is need to extract structured data (such as tables) from HTML content.
+                    2. `wikipedia` Python package is provided to interact with Wikipedia. Try to work with raw wikipedia HTML content and use `pandas` to parse first.
+                    3. `chess` Python package is provided. Please use it when there is need to solve chess problems.
+                    4. Please take the question literally! Do not add any additional information or assumptions.
+               
+            """,
+            verbosity_level=0,
+            max_steps=10,
+        )
+
+        self.manager_agent = CodeAgent(
+            model=InferenceClientModel(
+                "Qwen/Qwen2.5-32B-Instruct"
+            ),
+            tools=[get_file, VisitWebpageTool(), DuckDuckGoSearchTool()],
+            managed_agents=[
+                self.multimodal_agent, 
+                self.code_agent],
             additional_authorized_imports=[
                 "requests",
                 "bs4",
@@ -180,17 +167,18 @@ class BasicAgent:
                 "cv2",
                 "numpy",
                 "chess.engine",
-                "html5lib"
             ],
+            planning_interval=5,
             max_steps=15,
         )
 
-    def __call__(self, question: str, question_id: str, has_file: bool) -> str:
+    def __call__(self, question: str, question_id: str, file_name: str) -> str:
         print(f"Agent received question: {question}")
+        file = f"Mentioned file: {file_name}" if file_name else ""
         prompt = f"""
-            Answer the following question: 
-               "{question} {"The file name for this question is: " if has_file else ""} {question_id if has_file else ""}"
-             Please follow hints below:
+            Answer the following question (question_id is {question_id}):): 
+               "{question}""{file}"
+            Please follow hints below:
                 1. `pandas` Python package is provided. Please try to use it FIRST when there is need to extract structured data (such as tables) from HTML content.
                 2. `wikipedia` Python package is provided to interact with Wikipedia. Try to work with raw wikipedia HTML content and use `pandas` to parse first.
                 3. `chess` Python package is provided. Please use it when there is need to solve chess problems.