Update app.py

app.py

@@ -13,6 +13,9 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
 from transformers.models.llama.configuration_llama import LlamaConfig
 from huggingface_hub import hf_hub_download
 import gradio as gr
+from pathlib import Path
+
+
 
 # Solve permission issues
 os.environ["MPLCONFIGDIR"] = "/tmp/matplotlib"
@@ -29,7 +32,9 @@ os.environ["HF_HUB_DOWNLOAD_TIMEOUT"] = "60"
 os.environ["LANGCHAIN_TRACING_V2"] = "true"
 os.environ["LANGCHAIN_API_KEY"] = os.getenv("LANGCHAIN_API_KEY")
 os.environ["LANGCHAIN_PROJECT"] = os.getenv("LANGCHAIN_PROJECT")
-
+TEMP_DIR = tempfile.mkdtemp()
+os.environ['AUTOGEN_CACHE_DIR'] = TEMP_DIR
+os.environ['AUTOGEN_USE_DOCKER'] = "False"
 
 
 
@@ -1285,13 +1290,6 @@ def parse_query(query: str) -> dict:
 
 def autogen_multi_document_analysis(query: str, docs: list, file_names: list) -> str:
     try:
-        import os
-        import tempfile
-        
-        # 創建臨時目錄作為緩存目錄
-        cache_dir = tempfile.mkdtemp()
-        os.environ['AUTOGEN_CACHE_DIR'] = cache_dir
-        
         # 準備文件上下文
         context = "\n\n".join(
             f"Document {name}:\n{doc[:2000]}..." 
@@ -1304,70 +1302,76 @@ def autogen_multi_document_analysis(query: str, docs: list, file_names: list) ->
             "api_key": openai_api_key
         }]
 
+        # 創建臨時工作目錄
+        work_dir = os.path.join(TEMP_DIR, 'workdir')
+        os.makedirs(work_dir, exist_ok=True)
+
         # 基礎配置
         llm_config = {
             "config_list": config_list,
             "temperature": 0,
             "request_timeout": 120,
-            "seed": 42,
-            "cache_dir": cache_dir  # 使用臨時目錄作為緩存
+            "cache_seed": None  # 禁用緩存
         }
 
-        # 代碼執行配置
-        code_execution_config = {
+        # 配置執行設置
+        execution_config = {
             "use_docker": False,
-            "last_n_messages": 3,
+            "work_dir": work_dir,
             "timeout": 60,
-            "work_dir": cache_dir  # 使用臨時目錄作為工作目錄
+            "last_n_messages": 3
         }
 
-        # 定義代理人
+        # 定義用戶代理
         user_proxy = UserProxyAgent(
             name="User",
             system_message="A user seeking information from multiple documents.",
             human_input_mode="NEVER",
-            code_execution_config=code_execution_config
+            code_execution_config=execution_config,
+            llm_config=llm_config
         )
 
-        # 定義專家代理人們
-        expert_agents = [
-            AssistantAgent(
-                name="DocumentAnalyzer",
-                system_message="""You are an expert at analyzing documents. Your tasks include:
-                1. Comparing documents
-                2. Finding relationships
-                3. Identifying key differences and similarities
-                4. Analyzing project connections
-                Always provide evidence from the documents.""",
-                llm_config=llm_config
-            ),
-            AssistantAgent(
-                name="QAExpert",
-                system_message="""You are an expert at document QA. Your tasks include:
-                1. Answering specific questions
-                2. Finding relevant information
-                3. Cross-referencing content
-                Always cite specific parts of the documents.""",
-                llm_config=llm_config
-            ),
-            AssistantAgent(
-                name="Summarizer",
-                system_message="""You are an expert at summarizing. Your tasks include:
-                1. Creating comprehensive summaries
-                2. Identifying main points
-                3. Highlighting key findings
-                Focus on clarity and completeness.""",
-                llm_config=llm_config
-            )
-        ]
+        # 定義文檔分析專家
+        doc_analyzer = AssistantAgent(
+            name="DocumentAnalyzer",
+            system_message="""You are an expert at analyzing and comparing documents. Focus on:
+            1. Key similarities and differences
+            2. Main themes and topics
+            3. Relationships between documents
+            4. Evidence-based analysis""",
+            llm_config=llm_config
+        )
+
+        # 定義問答專家
+        qa_expert = AssistantAgent(
+            name="QAExpert",
+            system_message="""You are an expert at extracting specific information. Focus on:
+            1. Finding relevant details
+            2. Answering specific questions
+            3. Cross-referencing information
+            4. Providing evidence""",
+            llm_config=llm_config
+        )
+
+        # 定義總結專家
+        summarizer = AssistantAgent(
+            name="Summarizer",
+            system_message="""You are an expert at summarizing content. Focus on:
+            1. Key points and findings
+            2. Important relationships
+            3. Critical conclusions
+            4. Comprehensive overview""",
+            llm_config=llm_config
+        )
 
         # 創建群組聊天
         groupchat = GroupChat(
-            agents=[user_proxy] + expert_agents,
+            agents=[user_proxy, doc_analyzer, qa_expert, summarizer],
             messages=[],
             max_round=5
         )
 
+        # 創建管理器
         manager = GroupChatManager(
             groupchat=groupchat,
             llm_config=llm_config
@@ -1382,31 +1386,27 @@ def autogen_multi_document_analysis(query: str, docs: list, file_names: list) ->
         {context}
 
         Requirements:
-        1. Provide direct answers to the query
-        2. Support answers with evidence from documents
+        1. Provide a direct and clear answer
+        2. Support all claims with evidence from the documents
         3. Consider relationships between all documents
-        4. If comparing, analyze all relevant documents
-        5. If summarizing, consider all relevant content
-        6. If looking for specific content, search across all documents
-        7. If analyzing project relationships, check all possible connections
+        4. If comparing, analyze all relevant aspects
+        5. If summarizing, cover all important points
+        6. If looking for specific content, search thoroughly
+        7. If analyzing relationships, consider all connections
 
-        Please collaborate to provide a comprehensive answer."""
+        Please provide a comprehensive and well-structured answer."""
 
         # 執行群組討論
         user_proxy.initiate_chat(manager, message=task_prompt)
         result = user_proxy.last_message()["content"]
-        
-        # 清理臨時目錄
-        import shutil
-        try:
-            shutil.rmtree(cache_dir)
-        except:
-            pass
-            
+
         return result
 
     except Exception as e:
         print(f"ERROR in AutoGen processing: {str(e)}")
+        # 提供更有用的錯誤信息
+        if "Cache directory" in str(e):
+            return "Error: Unable to process the documents due to a system configuration issue. The analysis will be performed without caching."
         return f"Error during multi-agent processing: {str(e)}"