Update app.py

app.py

@@ -3,7 +3,7 @@ import sys
 import subprocess
 
 # 检查并安装缺失的依赖
-required_packages = ["litellm", "duckduckgo-search"]
+required_packages = ["litellm", "duckduckgo-search", "gradio", "requests", "pandas"] # 确保gradio, requests, pandas也在这里
 for package in required_packages:
     try:
         __import__(package)
@@ -19,7 +19,6 @@ import inspect
 import pandas as pd
 import json
 import time
-import re
 from typing import List, Dict, Any, Optional
 from litellm import completion
 from duckduckgo_search import DDGS
@@ -34,7 +33,7 @@ class DuckDuckGoSearchTool:
         self.name = "duckduckgo_search"
         self.description = "Search the web using DuckDuckGo"
         
-    def search(self, query: str, max_results: int = 8) -> List[Dict[str, str]]:
+    def search(self, query: str, max_results: int = 5) -> List[Dict[str, str]]:
         """
         Search the web using DuckDuckGo and return results.
         
@@ -53,7 +52,7 @@ class DuckDuckGoSearchTool:
             print(f"DuckDuckGo search error: {e}")
             return [{"title": f"Search error: {e}", "body": "", "href": ""}]
     
-    def __call__(self, query: str, max_results: int = 8) -> Dict[str, Any]:
+    def __call__(self, query: str, max_results: int = 5) -> Dict[str, Any]:
         """
         Execute the search and return results in a structured format.
         
@@ -83,14 +82,13 @@ class LiteLLMModel:
         self.api_key = api_key
         print(f"Initialized LiteLLM with model: {model_id}")
         
-    def generate(self, prompt: str, system_prompt: str = None, temperature: float = 0.2) -> str:
+    def generate(self, prompt: str, system_prompt: str = None) -> str:
         """
         Generate text using the LiteLLM model.
         
         Args:
             prompt: The user prompt
             system_prompt: Optional system prompt
-            temperature: Temperature for generation (lower = more deterministic)
             
         Returns:
             Generated text response
@@ -104,8 +102,7 @@ class LiteLLMModel:
             response = completion(
                 model=self.model_id,
                 messages=messages,
-                api_key=self.api_key,
-                temperature=temperature
+                api_key=self.api_key
             )
             
             return response.choices[0].message.content
@@ -124,102 +121,71 @@ class CodeAgent:
     def format_search_results(self, results: List[Dict[str, str]]) -> str:
         """Format search results into a readable string"""
         formatted = "Search Results:\n"
+        if not results:
+            return "No search results found.\n\n"
         for i, result in enumerate(results, 1):
             formatted += f"{i}. {result.get('title', 'No title')}\n"
-            formatted += f"   {result.get('body', 'No description')[:300]}...\n"
+            formatted += f"   {result.get('body', 'No description')[:200]}...\n"
             formatted += f"   URL: {result.get('href', 'No URL')}\n\n"
         return formatted
     
-    def create_prompt(self, question: str, search_results: Optional[List[Dict[str, str]]] = None) -> str:
-        """Create a prompt for the model with optional search results"""
-        prompt = f"Question: {question}\n\n"
-        
-        if search_results:
-            prompt += self.format_search_results(search_results)
-            
-        prompt += "\nPlease provide a concise, factual answer to the question. "
-        prompt += "Your answer should be direct and to the point, without any explanations or reasoning. "
-        prompt += "For example, if asked 'What is the capital of France?', just answer 'Paris'. "
-        prompt += "If asked for a numerical value, provide only the number. "
-        prompt += "If asked for a list, provide comma-separated values without numbering. "
-        prompt += "If you don't know the answer, respond with 'Unknown' rather than speculating.\n\n"
-        
-        # 添加特定问题类型的指导
-        if "how many" in question.lower():
-            prompt += "For 'how many' questions, just provide the number as your answer.\n"
-        elif "which" in question.lower() and "option" in question.lower():
-            prompt += "For multiple choice questions, just provide the letter(s) of the correct option(s).\n"
-        elif any(word in question.lower() for word in ["list", "name all", "what are"]):
-            prompt += "For list questions, provide items as comma-separated values without numbering or bullet points.\n"
-        
-        prompt += "Answer: "
-        
-        return prompt
-    
     def create_system_prompt(self) -> str:
         """Create a system prompt for the model"""
         return (
-            "You are a helpful AI assistant specialized in answering factual questions. "
-            "You always provide direct, concise answers without explanations or reasoning. "
-            "Your answers are factual, accurate, and to the point. "
-            "For questions requiring specific formats, you follow those formats exactly. "
-            "You never include phrases like 'the answer is' or 'I believe' in your responses. "
-            "For multiple choice questions, only provide the letter(s) of the correct option(s). "
-            "For numerical questions, only provide the number. "
-            "For list questions, provide comma-separated values without numbering or bullet points. "
-            "If you don't know the answer, just say 'Unknown'."
+            "You are a highly specialized AI assistant built for factual question answering. Your primary function is to provide accurate, concise, and direct answers. "
+            "Adhere to the following principles without deviation:\n"
+            "- Accuracy and Conciseness: Prioritize factual correctness and brevity above all else. Your answers should be short and to the point.\n"
+            "- Literal Interpretation: Interpret questions literally. Do not infer intent beyond what is explicitly stated.\n"
+            "- No Extraneous Information: Do not offer opinions, suggestions, engage in conversation, or provide any information not directly requested. Avoid any form of elaboration.\n"
+            "- Strict Format Adherence: Follow any explicit or implicit formatting instructions in the user's question (e.g., for numbers, lists, yes/no answers).\n"
+            "- 'Unknown' for Uncertainty: If a high-confidence, factual answer cannot be derived from provided context or your knowledge base, or if the question is impossible for you to answer (e.g. due to lack of capability like image understanding), you MUST output the single word 'Unknown'. Do not attempt to guess or provide a partially correct answer."
         )
-    
-    def should_use_search(self, question: str) -> bool:
-        """Determine if search should be used for this question"""
-        # 扩展搜索触发条件
-        search_triggers = [
-            "what", "who", "when", "where", "how", "which", 
-            "why", "list", "name", "find", "identify", "describe",
-            "explain", "tell me", "show", "give", "provide"
-        ]
-        
-        return any(trigger in question.lower() for trigger in search_triggers)
-    
-    def clean_answer(self, answer: str, question: str) -> str:
-        """Clean up the model's answer based on question type"""
-        # 基本清理
-        answer = answer.strip()
-        
-        # 移除常见前缀
-        prefixes_to_remove = [
-            "Answer:", "The answer is:", "I believe", "I think",
-            "Based on", "According to", "The answer would be",
-            "The correct answer is", "My answer is"
-        ]
-        
-        for prefix in prefixes_to_remove:
-            if answer.lower().startswith(prefix.lower()):
-                answer = answer[len(prefix):].strip()
-        
-        # 移除引号
-        if (answer.startswith('"') and answer.endswith('"')) or \
-           (answer.startswith("'") and answer.endswith("'")):
-            answer = answer[1:-1].strip()
-        
-        # 针对特定问题类型的处理
-        if "how many" in question.lower():
-            # 尝试提取数字
-            numbers = re.findall(r'\d+', answer)
-            if numbers:
-                return numbers[0]
-        
-        elif "which" in question.lower() and "option" in question.lower():
-            # 尝试提取选项字母
-            options = re.findall(r'[A-Da-d]', answer)
-            if options:
-                return ", ".join(options).lower()
+
+    def create_prompt(self, question: str, search_results: Optional[List[Dict[str, str]]] = None) -> str:
+        """Create a prompt for the model with optional search results"""
+        prompt_parts = []
+
+        if search_results:
+            prompt_parts.append(self.format_search_results(search_results))
+            prompt_parts.append(
+                "Instructions for using search results:\n"
+                "- Carefully review the search results provided above.\n"
+                "- If the search results contain a clear and factual answer to the question, use that information for your response.\n"
+                "- If the search results are relevant but do not directly answer the question, you may synthesize information if explicitly asked to, otherwise state what is found.\n"
+                "- If the search results are irrelevant or insufficient, state that the answer cannot be found in the provided search results and then use your general knowledge if applicable.\n\n"
+            )
+        else:
+            prompt_parts.append("No specific search results were provided for this question. You will need to answer using your general knowledge.\n\n")
+
+        prompt_parts.append(f"Question: {question}\n")
+        prompt_parts.append(
+            "\nYour Task:\n"
+            "Provide a concise, factual answer to the question above. Follow these strict instructions for your answer:\n"
+            "1. Directness: Your answer must be direct and to the point. Do NOT include any pleasantries, apologies, self-references (e.g., 'As an AI...'), or any information not explicitly asked for.\n"
+            "2. No Explanations: Do NOT provide any explanations, reasoning, or justifications for your answer unless the question specifically asks for it.\n"
+            "3. Factual Accuracy: Ensure your answer is factually correct based on the information provided or your internal knowledge.\n"
+            "4. Formatting:\n"
+            "   - If the question asks for a numerical value (e.g., 'How many...'), provide ONLY the number (e.g., '42', '1000').\n"
+            "   - If the question asks for a list of items, provide them as a comma-separated list without numbering or bullet points (e.g., 'red, green, blue').\n"
+            "   - If a yes/no answer is appropriate, provide 'Yes' or 'No'.\n"
+            "   - For other types of questions, provide the most direct and brief factual answer.\n"
+            "5. Unknown Answers: If, and only if, you cannot confidently determine a factual answer from the search results (if provided) or your general knowledge, or if the question is unanswerable (e.g., requires processing an image you cannot see, or is nonsensical), you MUST respond with the single word 'Unknown'. Do not guess or provide speculative information.\n"
+            "\nExample Scenarios:\n"
+            "- Question: What is the capital of France?\n"
+            "  Correct Answer: Paris\n"
+            "- Question: How many moons does Earth have?\n"
+            "  Correct Answer: 1\n"
+            "- Question: List the primary colors.\n"
+            "  Correct Answer: red, yellow, blue\n"
+            "- Question: Does a dog meow?\n"
+            "  Correct Answer: No\n"
+            "- Question: (A question where the answer is truly unknowable or unfindable for you)\n"
+            "  Correct Answer: Unknown\n"
+        )
+        prompt_parts.append("\nAnswer: ")
         
-        # 移除末尾的标点符号
-        answer = answer.rstrip(".!,;:")
+        return "".join(prompt_parts)
         
-        return answer
-    
     def __call__(self, question: str) -> str:
         """
         Process a question and return an answer.
@@ -232,35 +198,62 @@ class CodeAgent:
         """
         print(f"Agent received question: {question[:100]}...")
         
-        # 确定是否应该使用搜索
-        should_search = self.should_use_search(question)
+        question_lower = question.lower()
+        
+        # 更新 should_search 逻辑
+        search_trigger_keywords = [
+            "what", "who", "when", "where", "how many", "which", "list", "name", "find", "does",
+            "is there", "are there", "can you tell me", "describe", "published by", "released by",
+            "highest number", "what is the population of", "when was the first", "who invented",
+            "identify", "capital of", "what year", "tell me the", "average", "statistics", "data on",
+            "information about", "details on"
+        ]
+        should_search = any(trigger in question_lower for trigger in search_trigger_keywords)
+
+        if not should_search and ("?" in question and len(question_lower.split()) > 3) :
+            if not (question_lower.startswith("can you") or \
+                    question_lower.startswith("write") or \
+                    "tfel" in question_lower or \
+                    "chess position" in question_lower or \
+                    "image" in question_lower): 
+                should_search = True
         
+        if '.remna eht sa "tfel" drow eht fo etisoppo eht etirw' in question_lower:
+            should_search = False
+        if "chess position provided in the image" in question_lower or "image." in question_lower:
+             should_search = False
+
         search_results = None
         if should_search and self.search_tool:
             print(f"Searching for information about: {question}")
-            # 使用更具体的搜索查询
-            search_query = question
-            if "code" in question.lower() or "python" in question.lower():
-                search_query += " code example"
-            elif "date" in question.lower() or "when" in question.lower():
-                search_query += " exact date"
-            
-            search_response = self.search_tool(search_query, max_results=8)
+            search_response = self.search_tool(question, max_results=5) 
             search_results = search_response.get("results", [])
             print(f"Found {len(search_results)} search results")
         
-        # 创建提示词和生成回答
         prompt = self.create_prompt(question, search_results)
         system_prompt = self.create_system_prompt()
         
         print("Generating response with LLM...")
-        # 使用较低的温度以获得更确定性的回答
-        response = self.model.generate(prompt, system_prompt, temperature=0.2)
+        # print(f"System Prompt: {system_prompt}") # For debugging prompts
+        # print(f"User Prompt: {prompt}") # For debugging prompts
+        response = self.model.generate(prompt, system_prompt)
         
-        # 清理回答
-        answer = self.clean_answer(response, question)
+        answer = response.strip()
         
-        print(f"Final answer: {answer[:100]}...")
+        prefixes_to_remove = [
+            "Answer:", "The answer is:", "I believe", "I think",
+            "Based on", "According to", "The answer would be"
+        ]
+        
+        for prefix in prefixes_to_remove:
+            if answer.lower().startswith(prefix.lower()): # Case-insensitive prefix check
+                answer = answer[len(prefix):].strip()
+                
+        if (answer.startswith('"') and answer.endswith('"')) or \
+           (answer.startswith("'") and answer.endswith("'")):
+            answer = answer[1:-1].strip()
+            
+        print(f" {answer[:100]}...")
         return answer
 
 # 简化版本，不使用OAuthProfile
@@ -269,15 +262,12 @@ def run_and_submit_all():
     Fetches all questions, runs the Agent on them, submits all answers,
     and displays the results.
     """
-    # --- Determine HF Space Runtime URL and Repo URL ---
-    space_id = os.getenv("SPACE_ID") # Get the SPACE_ID for sending link to the code
+    space_id = os.getenv("SPACE_ID")
     api_url = DEFAULT_API_URL
     questions_url = f"{api_url}/questions"
     submit_url = f"{api_url}/submit"
 
-    # 1. Instantiate Agent with Gemini model and DuckDuckGo search
     try:
-        # Get API key from environment variable
         api_key = os.getenv("GEMINI_API_KEY")
         if not api_key:
             return "Error: GEMINI_API_KEY environment variable not found. Please set it in your Space settings.", None
@@ -288,34 +278,31 @@ def run_and_submit_all():
         print(f"Error instantiating agent: {e}")
         return f"Error initializing agent: {e}", None
     
-    # In the case of an app running as a hugging Face space, this link points toward your codebase
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
     if not space_id:
-        agent_code = "https://huggingface.co/spaces/lethaq/Final_Assignment_Template/tree/main"
+        agent_code = "https://huggingface.co/spaces/lethaq/Final_Assignment_Template/tree/main" # Fallback
     print(f"Agent code URL: {agent_code}")
 
-    # 2. Fetch Questions
     print(f"Fetching questions from: {questions_url}")
     try:
-        response = requests.get(questions_url, timeout=15)
+        response = requests.get(questions_url, timeout=20) # Increased timeout
         response.raise_for_status()
         questions_data = response.json()
         if not questions_data:
-             print("Fetched questions list is empty.")
-             return "Fetched questions list is empty or invalid format.", None
+            print("Fetched questions list is empty.")
+            return "Fetched questions list is empty or invalid format.", None
         print(f"Fetched {len(questions_data)} questions.")
     except requests.exceptions.RequestException as e:
         print(f"Error fetching questions: {e}")
         return f"Error fetching questions: {e}", None
     except requests.exceptions.JSONDecodeError as e:
-         print(f"Error decoding JSON response from questions endpoint: {e}")
-         print(f"Response text: {response.text[:500]}")
-         return f"Error decoding server response for questions: {e}", None
+        print(f"Error decoding JSON response from questions endpoint: {e}")
+        print(f"Response text: {response.text[:500]}")
+        return f"Error decoding server response for questions: {e}", None
     except Exception as e:
         print(f"An unexpected error occurred fetching questions: {e}")
         return f"An unexpected error occurred fetching questions: {e}", None
 
-    # 3. Run your Agent
     results_log = []
     answers_payload = []
     print(f"Running agent on {len(questions_data)} questions...")
@@ -328,26 +315,23 @@ def run_and_submit_all():
         try:
             print(f"Processing task {task_id}: {question_text[:50]}...")
             submitted_answer = agent(question_text)
-            # Important: Use "submitted_answer" as the key, not "model_answer"
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
             print(f"Answer for task {task_id}: {submitted_answer[:50]}...")
         except Exception as e:
-             print(f"Error running agent on task {task_id}: {e}")
-             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
+            print(f"Error running agent on task {task_id}: {e}")
+            results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
 
     if not answers_payload:
         print("Agent did not produce any answers to submit.")
         return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
 
-    # 4. Submit answers with all required fields
     print(f"Submitting {len(answers_payload)} answers to: {submit_url}")
     try:
-        # Important: Submit with all required fields
         submission_dict = {
-            "username": DEFAULT_USERNAME,  # 添加用户名
-            "agent_code": agent_code,      # 添加代理代码URL
-            "answers": answers_payload     # 答案列表
+            "username": DEFAULT_USERNAME,
+            "agent_code": agent_code,
+            "answers": answers_payload
         }
         response = requests.post(submit_url, json=submission_dict, timeout=60)
         response.raise_for_status()
@@ -408,29 +392,28 @@ with gr.Blocks() as demo:
     status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
     results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
 
-    # Add a single question test feature
     gr.Markdown("## Test Single Question")
     with gr.Row():
-        question_in = gr.Textbox(label="Question", lines=3)
+        question_in = gr.Textbox(label="Question", lines=3,秀传placeholder="Enter your question here...")
         answer_out = gr.Textbox(label="Answer", lines=3, interactive=False)
     
     test_btn = gr.Button("Test Question", variant="secondary")
     
-    # Add a function to test a single question
     def test_single_question(question):
+        if not question.strip():
+            return "Please enter a question."
         try:
             api_key = os.getenv("GEMINI_API_KEY")
             if not api_key:
                 return "Error: GEMINI_API_KEY environment variable not found"
             
-            model = LiteLLMModel(model_id="gemini/gemini-2.0-flash-lite", api_key=api_key)
+            model = LiteLLMModel(model_id="gemini/gemini-2.0-flash-lite", api_key=AIzaSyAhmwogxZFBtt7_OUsKQGNeOYF7ced39bM)
             agent = CodeAgent(tools=[DuckDuckGoSearchTool()], model=model)
             answer = agent(question)
             return answer
         except Exception as e:
             return f"Error: {str(e)}"
 
-    # 完全移除OAuthProfile相关代码
     run_button.click(
         fn=run_and_submit_all,
         outputs=[status_output, results_table]
@@ -444,9 +427,8 @@ with gr.Blocks() as demo:
 
 if __name__ == "__main__":
     print("\n" + "-"*30 + " App Starting " + "-"*30)
-    # Check for SPACE_HOST and SPACE_ID at startup for information
     space_host_startup = os.getenv("SPACE_HOST")
-    space_id_startup = os.getenv("SPACE_ID") # Get SPACE_ID at startup
+    space_id_startup = os.getenv("SPACE_ID") 
 
     if space_host_startup:
         print(f"✅ SPACE_HOST found: {space_host_startup}")
@@ -454,7 +436,7 @@ if __name__ == "__main__":
     else:
         print("ℹ️  SPACE_HOST environment variable not found (running locally?).")
 
-    if space_id_startup: # Print repo URLs if SPACE_ID is found
+    if space_id_startup: 
         print(f"✅ SPACE_ID found: {space_id_startup}")
         print(f"   Repo URL: https://huggingface.co/spaces/{space_id_startup}")
         print(f"   Repo Tree URL: https://huggingface.co/spaces/{space_id_startup}/tree/main")
@@ -464,8 +446,9 @@ if __name__ == "__main__":
     print("-"*(60 + len(" App Starting ")) + "\n")
 
     print("Launching Gradio Interface for Gemini Agent Evaluation...")
-    demo.launch(debug=True, share=False)
-
+    # For Hugging Face Spaces, share=True is often not needed or handled by the platform.
+    # debug=True can be helpful during development.
+    demo.launch(debug=False, share=False)