Update app.py

app.py

@@ -4,102 +4,132 @@ import requests
 import pandas as pd
 import time
 from typing import Optional
-from smolagents import CodeAgent, OpenAIServerModel, tool
 
-# --- 搜尋工具 (修正安裝指令 + 省流量版) ---
+# 引入搜尋工具
 try:
     from duckduckgo_search import DDGS
 except ImportError:
     import os
-    # 修正：使用正確的套件名稱進行安裝
-    os.system('pip install duckduckgo-search')
+    os.system('pip install duckduckgo-search>=6.0.0')
     from duckduckgo_search import DDGS
 
-@tool
-def web_search(query: str) -> str:
-    """
-    Lite search with strict token limits for Groq free tier.
-    Args:
-        query: The search query string.
-    """
-    print(f"🕵️ [Debug] Searching: {query}")
+DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
+GROQ_API_URL = "https://api.groq.com/openai/v1/chat/completions"
+
+def perform_search(query: str) -> str:
+    """手動執行搜尋並回傳摘要"""
+    print(f"🕵️ Searching for: {query}")
     try:
-        # 強制冷靜 1 秒，保護新 Key
-        time.sleep(1.0)
-        
+        # 為了避免被封鎖，加一點延遲
+        time.sleep(1)
         with DDGS() as ddgs:
-            # 使用 lite 模式，只抓 1 筆
-            results = list(ddgs.text(query, max_results=1, backend="lite"))
+            # 使用 lite 模式比較快且穩
+            results = list(ddgs.text(query, max_results=3, backend="lite"))
         
         if not results:
-            return "No results. Guess answer."
+            return "No search results found."
             
-        # 關鍵保命符：截斷內容！
-        # 只回傳前 400 字，避免瞬間 Token 爆量導致卡死
-        first = results[0]
-        summary = f"Title: {first.get('title','')}\nInfo: {first.get('body','')}"
-        return summary[:400] 
+        # 整理結果
+        context = []
+        for r in results:
+            context.append(f"- {r.get('body', '')}")
         
+        # 限制長度以免爆 Token
+        return "\n".join(context)[:1000]
     except Exception as e:
-        print(f"❌ Error: {e}")
-        return "Search error. Guess answer."
-
-# -----------------------------------------------------------
-
-DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
+        print(f"Search Error: {e}")
+        return "Search failed."
 
 class GroqAgent:
+    """使用 Groq API + 手動搜尋的 Agent"""
     def __init__(self):
         self.api_key = os.getenv("GROQ_API_KEY")
         if not self.api_key:
+            print("✗ GROQ_API_KEY not found")
             self.agent = None
             return
-            
-        # 使用 Llama 3.3 70B
-        model = OpenAIServerModel(
-            model_id="llama-3.3-70b-versatile",
-            api_base="https://api.groq.com/openai/v1",
-            api_key=self.api_key
-        )
         
-        self.agent = CodeAgent(
-            tools=[web_search], 
-            model=model,
-            max_steps=2, 
-            verbosity_level=1
-        )
+        self.agent = True
+        print("✓ Groq agent initialized")
     
     def __call__(self, question: str) -> str:
         if self.agent is None:
             return "Error: GROQ_API_KEY not configured."
         
         try:
-            # 強制冷靜 2 秒，避免觸發 RPM 限制
-            time.sleep(2.0)
+            # 1. 先進行搜尋 (這是拿分的關鍵！)
+            # 我們直接搜整個問題，或者你可以寫邏輯去提取關鍵字
+            search_context = perform_search(question)
+            
+            # 2. 組合新的 Prompt
+            system_prompt = """
+            You are a helpful AI assistant. 
+            You will be provided with Context from a web search.
+            Use the Context to answer the User's Question accurately.
+            If the answer is in the Context, use it.
+            If the Context is empty or irrelevant, use your internal knowledge.
+            Keep answers concise.
+            """
             
-            prompt = f"""
-            Task: Answer concisely.
-            1. Search ONCE.
-            2. If search fails, GUESS.
-            3. Keep answer under 20 words.
+            user_content = f"""
+            Context:
+            {search_context}
             
-            Question: {question}
+            Question:
+            {question}
             """
-            return str(self.agent.run(prompt))
+
+            # 3. 呼叫 Groq
+            headers = {
+                "Authorization": f"Bearer {self.api_key}",
+                "Content-Type": "application/json"
+            }
+            
+            payload = {
+                "model": "llama-3.3-70b-versatile", # 70B 比較聰明
+                "messages": [
+                    {"role": "system", "content": system_prompt},
+                    {"role": "user", "content": user_content}
+                ],
+                "temperature": 0.1, # 降低隨機性
+                "max_tokens": 300,
+            }
+            
+            # 強制冷靜一下避免 Rate Limit
+            time.sleep(1)
+            
+            response = requests.post(
+                GROQ_API_URL,
+                headers=headers,
+                json=payload,
+                timeout=30
+            )
+            
+            if response.status_code != 200:
+                return f"API Error {response.status_code}"
+            
+            result = response.json()
+            answer = result['choices'][0]['message']['content'].strip()
+            
+            return answer
+            
         except Exception as e:
             return f"Error: {str(e)[:100]}"
 
+
 def run_and_submit_all(profile: Optional[gr.OAuthProfile] = None):
+    """主要評估和提交函數"""
+    
     if profile is None:
-        return "⚠️ Please login first!", None
+        return "⚠️ Please click 'Login with Hugging Face' button first!", None
     
     username = profile.username
     space_id = os.getenv("SPACE_ID", "s1144662")
     api_url = DEFAULT_API_URL
     
     try:
-        agent_wrapper = GroqAgent()
-        if agent_wrapper.agent is None:
+        agent = GroqAgent()
+        if agent.agent is None:
             return "❌ Error: GROQ_API_KEY not found!", None
     except Exception as e:
         return f"❌ Init failed: {str(e)}", None
@@ -107,48 +137,72 @@ def run_and_submit_all(profile: Optional[gr.OAuthProfile] = None):
     try:
         print("Fetching questions...")
         response = requests.get(f"{api_url}/questions", timeout=30)
-        questions = response.json()
+        questions_data = response.json()
     except Exception as e:
-        return f"❌ Fetch failed: {str(e)}", None
+        return f"❌ Failed to fetch questions: {str(e)}", None
 
-    answers = []
-    logs = []
+    answers_payload = []
+    results_log = []
     
-    total = len(questions)
-    for idx, item in enumerate(questions, 1):
-        q = item.get("question")
-        tid = item.get("task_id")
-        
-        print(f"🚀 [{idx}/{total}] Task: {tid}")
+    total = len(questions_data)
+    
+    for idx, item in enumerate(questions_data, 1):
+        task_id = item.get("task_id")
+        question_text = item.get("question")
         
-        ans = agent_wrapper(q)
+        print(f"[{idx}/{total}] Processing: {task_id}...")
         
-        answers.append({"task_id": tid, "submitted_answer": ans})
-        logs.append({"Task": tid, "Answer": str(ans)[:100]})
+        try:
+            answer = agent(question_text)
+            
+            answers_payload.append({
+                "task_id": task_id,
+                "submitted_answer": answer
+            })
+            
+            results_log.append({
+                "Task ID": task_id,
+                "Question": question_text[:50],
+                "Answer": answer[:100]
+            })
+            
+        except Exception as e:
+            answers_payload.append({"task_id": task_id, "submitted_answer": "Error"})
+            results_log.append({"Task ID": task_id, "Question": "Error", "Answer": str(e)})
 
     try:
-        print("Submitting...")
-        res = requests.post(f"{api_url}/submit", json={
+        print("Submitting answers...")
+        submission_data = {
             "username": username,
             "agent_code": f"https://huggingface.co/spaces/{space_id}/tree/main",
-            "answers": answers
-        }, timeout=60)
+            "answers": answers_payload
+        }
         
-        data = res.json()
+        response = requests.post(
+            f"{api_url}/submit",
+            json=submission_data,
+            timeout=120
+        )
+        data = response.json()
         score = data.get('score', 0)
-        return f"🎉 Score: {score}%", pd.DataFrame(logs)
         
+        status_msg = f"🎉 Score: {score}%" if score >= 30 else f"Score: {score}% (Need 30%)"
+        return status_msg, pd.DataFrame(results_log)
+            
     except Exception as e:
-        return f"Submit error: {str(e)}", pd.DataFrame(logs)
+        return f"❌ Submission failed: {str(e)}", pd.DataFrame(results_log)
 
-with gr.Blocks(title="Final Agent (New Key + Token Safe)") as demo:
-    gr.Markdown("# 🚀 Final Agent (New Key)")
+# Gradio 介面
+with gr.Blocks(title="Unit 4 Final Assignment (Manual RAG)", theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# 🚀 Final Agent (Manual Search)")
     with gr.Row():
-        gr.LoginButton()
-        btn = gr.Button("Run Evaluation", variant="primary")
-    out = gr.Textbox(label="Status")
-    tab = gr.DataFrame(label="Logs")
-    btn.click(run_and_submit_all, outputs=[out, tab])
+        gr.LoginButton(scale=1)
+        run_btn = gr.Button("Run Evaluation", scale=3, variant="primary")
+    
+    status = gr.Textbox(label="Status", lines=2)
+    details = gr.DataFrame(label="Results")
+    
+    run_btn.click(fn=run_and_submit_all, inputs=[], outputs=[status, details])
 
 if __name__ == "__main__":
     demo.launch()