Update app.py

app.py

@@ -6,7 +6,8 @@ import pandas as pd
 import time
 import re
 from markdownify import markdownify
-from smolagents import Tool, DuckDuckGoSearchTool, CodeAgent, WikipediaSearchTool, LiteLLMModel
+from smolagents import Tool, DuckDuckGoSearchTool, CodeAgent, WikipediaSearchTool
+from langchain_anthropic import ChatAnthropic
 from datetime import datetime, timedelta
 import threading
 
@@ -14,10 +15,10 @@ import threading
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
-# Rate limiting configuration
-RATE_LIMIT_REQUESTS = 18  # Stay below the 20/min limit
+# Rate limiting configuration for Anthropic (more generous limits)
+RATE_LIMIT_REQUESTS = 50  # Anthropic has higher rate limits
 RATE_LIMIT_WINDOW = 60    # 60 seconds
-REQUEST_DELAY = 4         # Minimum delay between requests (60/18 ≈ 3.33, using 4 for safety)
+REQUEST_DELAY = 1         # Reduced delay since Anthropic has better rate limits
 
 class RateLimiter:
     def __init__(self, max_requests=RATE_LIMIT_REQUESTS, window_seconds=RATE_LIMIT_WINDOW):
@@ -109,17 +110,27 @@ class VisitWebpageTool(Tool):
     def __init__(self, *args, **kwargs):
         self.is_initialized = False
 
-# --- Improved Agent Definition ---
+# --- Custom Agent using Claude directly ---
 class BasicAgent:
     def __init__(self):
+        # Initialize Anthropic Claude model
+        API_KEY = os.getenv("ANTHROPIC_API_KEY")
+        if not API_KEY:
+            raise ValueError("ANTHROPIC_API_KEY not found in environment variables.")
+        
+        self.model_name = "claude-3-haiku-20240307"
+        self.chat_model = ChatAnthropic(model=self.model_name, anthropic_api_key=API_KEY)
         self.rate_limiter = RateLimiter()
-        self.agent = CodeAgent(
-            model=LiteLLMModel(model_id="openrouter/meta-llama/llama-4-maverick:free", api_key=os.getenv("OPENROUTER_KEY")),
-            tools=[DuckDuckGoSearchTool(), WikipediaSearchTool(), VisitWebpageTool(), DownloadTaskAttachmentTool()],
-            add_base_tools=True,
-            additional_authorized_imports=['pandas','numpy','csv','subprocess', 'exec']
-        )
-        print("BasicAgent initialized.")
+        
+        # Initialize tools
+        self.tools = {
+            'search': DuckDuckGoSearchTool(),
+            'wikipedia': WikipediaSearchTool(),
+            'webpage': VisitWebpageTool(),
+            'download': DownloadTaskAttachmentTool()
+        }
+        
+        print(f"BasicAgent initialized with Claude model: {self.model_name}")
     
     def __call__(self, question: str, max_retries: int = 3) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
@@ -129,9 +140,14 @@ class BasicAgent:
                 # Apply rate limiting
                 self.rate_limiter.wait_if_needed()
                 
-                # Run the agent
-                agent_answer = self.agent.run(question)
-                print(f"Agent returning answer: {agent_answer}")
+                # Create a comprehensive prompt for Claude
+                prompt = self._create_prompt(question)
+                
+                # Get response from Claude
+                response = self.chat_model.invoke(prompt)
+                agent_answer = response.content
+                
+                print(f"Agent returning answer: {agent_answer[:100]}...")
                 return agent_answer
                 
             except Exception as e:
@@ -152,6 +168,30 @@ class BasicAgent:
                     return f"AGENT_ERROR: {error_msg}"
         
         return "MAX_RETRIES_EXCEEDED"
+    
+    def _create_prompt(self, question: str) -> str:
+        """Create a comprehensive prompt for Claude to answer the question"""
+        prompt = f"""You are a helpful AI agent tasked with answering questions accurately and comprehensively. 
+
+You have access to the following tools if needed:
+- Web search for current information
+- Wikipedia search for factual information
+- Webpage visiting for detailed content
+- File downloading for task-specific files
+
+Question: {question}
+
+Please provide a clear, accurate, and comprehensive answer. If you need to use external tools or resources, describe what you would do, but provide your best direct answer based on your training data.
+
+If the question involves:
+- Current events or recent information: Mention that you would use web search
+- Specific factual lookups: Mention that you would use Wikipedia or web search
+- File analysis: Mention that you would download and analyze the file
+- Code or technical problems: Provide working solutions with explanations
+
+Answer:"""
+        
+        return prompt
 
     def download_file(self, task_id: str) -> str:
         """
@@ -196,7 +236,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, progress=gr.Progress()):
     submit_url = f"{api_url}/submit"
 
     # 1. Instantiate Agent
-    progress(0, desc="Initializing agent...")
+    progress(0, desc="Initializing Claude agent...")
     try:
         agent = BasicAgent()
     except Exception as e:
@@ -231,7 +271,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, progress=gr.Progress()):
     results_log = []
     answers_payload = []
     total_questions = len(questions_data)
-    print(f"Running agent on {total_questions} questions...")
+    print(f"Running Claude agent on {total_questions} questions...")
     
     for i, item in enumerate(questions_data):
         progress((0.1 + 0.8 * i / total_questions), desc=f"Processing question {i+1}/{total_questions}")
@@ -251,7 +291,15 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, progress=gr.Progress()):
             if requires_file:
                 file_path = agent.download_file(task_id)
                 print(f"File for task {task_id} saved at: {file_path}")
-                submitted_answer = agent(f"{question_text} (File: {file_path})")
+                # Read file content and include in question
+                try:
+                    with open(file_path, 'r', encoding='utf-8') as f:
+                        file_content = f.read()
+                    enhanced_question = f"{question_text}\n\nFile content:\n{file_content}"
+                except:
+                    # If can't read as text, just mention the file path
+                    enhanced_question = f"{question_text}\n\nFile downloaded to: {file_path}"
+                submitted_answer = agent(enhanced_question)
             else:
                 submitted_answer = agent(question_text)
 
@@ -280,7 +328,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, progress=gr.Progress()):
     # 4. Prepare Submission 
     progress(0.9, desc="Submitting answers...")
     submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
-    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
+    status_update = f"Claude agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
     print(status_update)
 
     # 5. Submit
@@ -296,6 +344,7 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, progress=gr.Progress()):
             f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
             f"Processed: {len(results_log)} questions\n"
             f"Successfully submitted: {len(answers_payload)} answers\n"
+            f"Model used: Claude 3 Haiku\n"
             f"Message: {result_data.get('message', 'No message received.')}"
         )
         print("Submission successful.")
@@ -331,24 +380,24 @@ def run_and_submit_all(profile: gr.OAuthProfile | None, progress=gr.Progress()):
 
 # --- Build Gradio Interface using Blocks ---
 with gr.Blocks() as demo:
-    gr.Markdown("# Basic Agent Evaluation Runner")
+    gr.Markdown("# Claude Agent Evaluation Runner")
     gr.Markdown(
         """
         **Instructions:**
 
         1. Please clone this space, then modify the code to define your agent's logic, the tools, the necessary packages, etc.
-        2. Log in to your Hugging Face account using the button below. This uses your HF username for submission.
-        3. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your agent, submit answers, and see the score.
+        2. Make sure you have set your `ANTHROPIC_API_KEY` environment variable.
+        3. Log in to your Hugging Face account using the button below. This uses your HF username for submission.
+        4. Click 'Run Evaluation & Submit All Answers' to fetch questions, run your Claude agent, submit answers, and see the score.
 
         ---
-        **Improvements:**
-        - ✅ Rate limiting to prevent API errors
-        - ✅ Progressive retry logic with backoff
-        - ✅ Better error handling and categorization
-        - ✅ Progress tracking during execution
-        - ✅ Detailed status reporting
+        **Model Configuration:**
+        - 🤖 Using Claude 3 Haiku via Anthropic API
+        - ⚡ Higher rate limits compared to free tier models
+        - 🛠️ Custom prompt engineering for better responses
+        - 📁 Enhanced file handling for task attachments
         
-        **Note:** This improved version includes rate limiting to stay within the free tier limits of 20 requests per minute.
+        **Note:** This version uses your Anthropic Claude model directly instead of smolagents CodeAgent.
         """
     )
 
@@ -367,6 +416,14 @@ with gr.Blocks() as demo:
 
 if __name__ == "__main__":
     print("\n" + "-"*30 + " App Starting " + "-"*30)
+    
+    # Check for required API key
+    api_key_check = os.getenv("ANTHROPIC_API_KEY")
+    if api_key_check:
+        print("✅ ANTHROPIC_API_KEY found")
+    else:
+        print("❌ ANTHROPIC_API_KEY not found - please set this environment variable")
+    
     space_host_startup = os.getenv("SPACE_HOST")
     space_id_startup = os.getenv("SPACE_ID")
 
@@ -385,5 +442,5 @@ if __name__ == "__main__":
 
     print("-"*(60 + len(" App Starting ")) + "\n")
 
-    print("Launching Gradio Interface for Basic Agent Evaluation...")
+    print("Launching Gradio Interface for Claude Agent Evaluation...")
     demo.launch(debug=True, share=False)