Merge branch 'main' of http://192.168.110.18:30000/admin/Project02

AITrain/dual_ai_dialogue_system.py

@@ -64,14 +64,39 @@ class RAGKnowledgeBase:
     
     def _load_knowledge_base(self):
         """加载知识库"""
-        # 加载世界观
-        worldview_files = [f for f in os.listdir(self.knowledge_dir) 
-                          if f.startswith('worldview') and f.endswith('.json')]
-        if worldview_files:
-            worldview_path = os.path.join(self.knowledge_dir, worldview_files[0])
-            with open(worldview_path, 'r', encoding='utf-8') as f:
-                self.worldview_data = json.load(f)
-            print(f"✓ 世界观加载成功: {self.worldview_data.get('worldview_name', '未知')}")
+        # 优先加载RAG知识库作为世界观
+        rag_worldview_path = "./rag_knowledge/knowledge_base.json"
+        if os.path.exists(rag_worldview_path):
+            try:
+                with open(rag_worldview_path, 'r', encoding='utf-8') as f:
+                    rag_data = json.load(f)
+                    # 从RAG数据中提取世界观信息
+                    self.worldview_data = {
+                        "worldview_name": "克苏鲁神话世界观 (RAG)",
+                        "source": rag_data.get("metadata", {}).get("source_file", "未知"),
+                        "description": f"基于{rag_data.get('metadata', {}).get('source_file', 'PDF文档')}的RAG知识库",
+                        "total_chunks": rag_data.get("metadata", {}).get("total_chunks", 0),
+                        "total_concepts": rag_data.get("metadata", {}).get("total_concepts", 0),
+                        "rag_enabled": True
+                    }
+                    # 保存RAG数据用于检索
+                    self.rag_chunks = rag_data.get("chunks", [])
+                    print(f"✓ RAG世界观加载成功: {self.worldview_data['worldview_name']}")
+                    print(f"  - 文档块数: {self.worldview_data['total_chunks']}")
+                    print(f"  - 概念数: {self.worldview_data['total_concepts']}")
+            except Exception as e:
+                print(f"✗ RAG世界观加载失败: {e}")
+                self.rag_chunks = []
+        
+        # 如果没有RAG知识库，则加载传统世界观文件
+        if not hasattr(self, 'rag_chunks') or not self.rag_chunks:
+            worldview_files = [f for f in os.listdir(self.knowledge_dir) 
+                              if f.startswith('worldview') and f.endswith('.json')]
+            if worldview_files:
+                worldview_path = os.path.join(self.knowledge_dir, worldview_files[0])
+                with open(worldview_path, 'r', encoding='utf-8') as f:
+                    self.worldview_data = json.load(f)
+                print(f"✓ 传统世界观加载成功: {self.worldview_data.get('worldview_name', '未知')}")
         
         # 加载角色数据
         character_files = [f for f in os.listdir(self.knowledge_dir) 
@@ -96,21 +121,38 @@ class RAGKnowledgeBase:
         """构建可检索的文本块"""
         self.chunks = []
         
-        # 世界观相关文本块
-        if self.worldview_data:
-            for section_key, section_data in self.worldview_data.items():
-                if isinstance(section_data, dict):
-                    for sub_key, sub_data in section_data.items():
-                        if isinstance(sub_data, (str, list)):
-                            content = str(sub_data)
-                            if len(content) > 50:  # 只保留有意义的文本
-                                self.chunks.append({
-                                    "type": "worldview",
-                                    "section": section_key,
-                                    "subsection": sub_key,
-                                    "content": content,
-                                    "metadata": {"source": "worldview"}
-                                })
+        # 优先使用RAG知识库的文本块
+        if hasattr(self, 'rag_chunks') and self.rag_chunks:
+            for rag_chunk in self.rag_chunks:
+                self.chunks.append({
+                    "type": "worldview_rag",
+                    "section": "rag_knowledge",
+                    "subsection": rag_chunk.get("type", "unknown"),
+                    "content": rag_chunk.get("content", ""),
+                    "metadata": {
+                        "source": "rag_worldview",
+                        "chunk_id": rag_chunk.get("id", ""),
+                        "size": rag_chunk.get("size", 0),
+                        "hash": rag_chunk.get("hash", "")
+                    }
+                })
+            print(f"✓ 使用RAG知识库文本块: {len(self.rag_chunks)} 个")
+        else:
+            # 传统世界观相关文本块
+            if self.worldview_data:
+                for section_key, section_data in self.worldview_data.items():
+                    if isinstance(section_data, dict):
+                        for sub_key, sub_data in section_data.items():
+                            if isinstance(sub_data, (str, list)):
+                                content = str(sub_data)
+                                if len(content) > 50:  # 只保留有意义的文本
+                                    self.chunks.append({
+                                        "type": "worldview",
+                                        "section": section_key,
+                                        "subsection": sub_key,
+                                        "content": content,
+                                        "metadata": {"source": "worldview"}
+                                    })
         
         # 角色相关文本块
         for char_name, char_data in self.character_data.items():
@@ -134,6 +176,18 @@ class RAGKnowledgeBase:
     def _build_vector_index(self):
         """构建向量索引"""
         try:
+            # 优先使用RAG知识库的预构建向量索引
+            rag_vector_path = "./rag_knowledge/vector_index.faiss"
+            rag_embeddings_path = "./rag_knowledge/embeddings.npy"
+            
+            if os.path.exists(rag_vector_path) and os.path.exists(rag_embeddings_path):
+                # 加载预构建的向量索引
+                self.index = faiss.read_index(rag_vector_path)
+                self.rag_embeddings = np.load(rag_embeddings_path)
+                print(f"✓ 使用RAG预构建向量索引: {self.index.ntotal}个向量")
+                return
+            
+            # 如果没有预构建的向量索引，则重新构建
             texts = [chunk["content"] for chunk in self.chunks]
             embeddings = self.embedding_model.encode(texts)
             
@@ -152,14 +206,26 @@ class RAGKnowledgeBase:
         # 向量搜索
         if EMBEDDING_AVAILABLE and self.embedding_model and self.index:
             try:
-                query_vector = self.embedding_model.encode([query])
-                distances, indices = self.index.search(query_vector.astype(np.float32), top_k * 2)
+                # 如果使用RAG预构建向量索引，直接搜索
+                if hasattr(self, 'rag_embeddings'):
+                    query_vector = self.embedding_model.encode([query])
+                    distances, indices = self.index.search(query_vector.astype(np.float32), top_k * 2)
                     
-                for distance, idx in zip(distances[0], indices[0]):
-                    if idx < len(self.chunks):
-                        chunk = self.chunks[idx].copy()
-                        chunk["relevance_score"] = float(1 / (1 + distance))
-                        relevant_chunks.append(chunk)
+                    for distance, idx in zip(distances[0], indices[0]):
+                        if idx < len(self.chunks):
+                            chunk = self.chunks[idx].copy()
+                            chunk["relevance_score"] = float(1 / (1 + distance))
+                            relevant_chunks.append(chunk)
+                else:
+                    # 传统向量搜索
+                    query_vector = self.embedding_model.encode([query])
+                    distances, indices = self.index.search(query_vector.astype(np.float32), top_k * 2)
+                    
+                    for distance, idx in zip(distances[0], indices[0]):
+                        if idx < len(self.chunks):
+                            chunk = self.chunks[idx].copy()
+                            chunk["relevance_score"] = float(1 / (1 + distance))
+                            relevant_chunks.append(chunk)
             except Exception as e:
                 print(f"向量搜索失败: {e}")
         
@@ -317,8 +383,17 @@ class DualAIDialogueEngine:
         self.conv_mgr = conversation_manager
         self.llm_generator = llm_generator
     
-    def generate_character_prompt(self, character_name: str, context_info: List[Dict], dialogue_history: List[DialogueTurn]) -> str:
-        """为角色生成对话提示"""
+    def generate_character_prompt(self, character_name: str, context_info: List[Dict], dialogue_history: List[DialogueTurn], 
+                                history_context_count: int = 3, context_info_count: int = 2) -> str:
+        """为角色生成对话提示
+        
+        Args:
+            character_name: 角色名称
+            context_info: 相关上下文信息
+            dialogue_history: 对话历史
+            history_context_count: 使用的历史对话轮数（默认3轮）
+            context_info_count: 使用的上下文信息数量（默认2个）
+        """
         char_data = self.kb.character_data.get(character_name, {})
         
         # 基础角色设定
@@ -338,56 +413,86 @@ class DualAIDialogueEngine:
             situation = char_data['current_situation']
             prompt_parts.append(f"当前状态：{situation.get('current_mood', '')}")
         
-        # 相关世界观信息
+        # 相关世界观信息（可控制数量）
         if context_info:
             prompt_parts.append("相关背景信息：")
-            for info in context_info[:2]:  # 只使用最相关的2个信息
+            for info in context_info[:context_info_count]:
                 content = info['content'][:200] + "..." if len(info['content']) > 200 else info['content']
                 prompt_parts.append(f"- {content}")
         
-        # 对话历史
+        # 对话历史（可控制数量）
         if dialogue_history:
             prompt_parts.append("最近的对话：")
-            for turn in dialogue_history[-3:]:  # 只使用最近的3轮对话
+            # 使用参数控制历史对话轮数
+            history_to_use = dialogue_history[-history_context_count:] if history_context_count > 0 else []
+            for turn in history_to_use:
                 prompt_parts.append(f"{turn.speaker}: {turn.content}")
         
         prompt_parts.append("\n请根据角色设定和上下文，生成符合角色特点的自然对话。回复应该在50-150字之间。")
         
         return "\n".join(prompt_parts)
     
-    def generate_dialogue(self, session_id: str, current_speaker: str, topic_hint: str = "") -> Tuple[str, List[str]]:
-        """生成角色对话"""
+    def generate_dialogue(self, session_id: str, current_speaker: str, topic_hint: str = "", 
+                         history_context_count: int = 3, context_info_count: int = 2) -> Tuple[str, List[str]]:
+        """生成角色对话
+        
+        Args:
+            session_id: 会话ID
+            current_speaker: 当前说话者
+            topic_hint: 话题提示
+            history_context_count: 使用的历史对话轮数（默认3轮）
+            context_info_count: 使用的上下文信息数量（默认2个）
+        """
         # 获取对话历史
         dialogue_history = self.conv_mgr.get_conversation_history(session_id)
         
         # 构建搜索查询
         if dialogue_history:
-            # 基于最近的对话内容
-            recent_content = " ".join([turn.content for turn in dialogue_history[-2:]])
+            # 基于最近的对话内容（可控制数量）
+            recent_turns = dialogue_history[-history_context_count:] if history_context_count > 0 else []
+            recent_content = " ".join([turn.content for turn in recent_turns])
             search_query = recent_content + " " + topic_hint
         else:
             # 首次对话
             search_query = f"{current_speaker} {topic_hint} introduction greeting"
         
         # 搜索相关上下文
-        context_info = self.kb.search_relevant_context(search_query, current_speaker,  10)
+        context_info = self.kb.search_relevant_context(search_query, current_speaker, context_info_count)
         
-        # 生成提示
-        prompt = self.generate_character_prompt(current_speaker, context_info, dialogue_history)
+        # 生成提示（使用参数控制上下文数量）
+        prompt = self.generate_character_prompt(
+            current_speaker, 
+            context_info, 
+            dialogue_history,
+            history_context_count,
+            context_info_count
+        )
         
-        # 生成对话
+        # 生成对话 - 使用双模型系统
         try:
-            response = self.llm_generator.generate_character_dialogue(
-                current_speaker, 
-                prompt, 
-                topic_hint or "请继续对话",
-                temperature=0.8,
-                max_new_tokens=150
-            )
+            # 检查是否为双模型对话系统
+            if hasattr(self.llm_generator, 'generate_dual_character_dialogue'):
+                # 使用双模型系统
+                response = self.llm_generator.generate_dual_character_dialogue(
+                    current_speaker, 
+                    prompt, 
+                    topic_hint or "请继续对话",
+                    temperature=0.8,
+                    max_new_tokens=150
+                )
+            else:
+                # 兼容旧的单模型系统
+                response = self.llm_generator.generate_character_dialogue(
+                    current_speaker, 
+                    prompt, 
+                    topic_hint or "请继续对话",
+                    temperature=0.8,
+                    max_new_tokens=150
+                )
             
             # 记录使用的上下文
-            context_used = [f"{info['section']}.{info['subsection']}" for info in context_info]
-            avg_relevance = sum(info['relevance_score'] for info in context_info) / len(context_info) if context_info else 0.0
+            context_used = [f"{info['section']}.{info['subsection']}" for info in context_info[:context_info_count]]
+            avg_relevance = sum(info['relevance_score'] for info in context_info[:context_info_count]) / len(context_info[:context_info_count]) if context_info else 0.0
             
             # 保存对话轮次
             self.conv_mgr.add_dialogue_turn(
@@ -400,191 +505,345 @@ class DualAIDialogueEngine:
             print(f"✗ 对话生成失败: {e}")
             return f"[{current_speaker}暂时无法回应]", []
     
-    def run_conversation_turn(self, session_id: str, characters: List[str], turns_count: int = 1, topic: str = ""):
-        """运行对话轮次"""
-        results = []
+    def run_conversation_turn(self, session_id: str, characters: List[str], turns_count: int = 1, topic: str = "",
+                             history_context_count: int = 3, context_info_count: int = 2):
+        """运行对话轮次
         
+        Args:
+            session_id: 会话ID
+            characters: 角色列表
+            turns_count: 对话轮数
+            topic: 对话主题
+            history_context_count: 使用的历史对话轮数（默认3轮）
+            context_info_count: 使用的上下文信息数量（默认2个）
+        """
+        results = []
+        print(f"  [上下文设置: 历史{history_context_count}轮, 信息{context_info_count}个]")
         for i in range(turns_count):
             for char in characters:
-                response, context_used = self.generate_dialogue(session_id, char, topic)
+                response, context_used = self.generate_dialogue(
+                    session_id, 
+                    char, 
+                    topic,
+                    history_context_count,
+                    context_info_count
+                )
                 results.append({
                     "speaker": char,
                     "content": response,
                     "context_used": context_used,
-                    "turn": i + 1
+                    "turn": i + 1,
+                    "context_settings": {
+                        "history_count": history_context_count,
+                        "context_info_count": context_info_count
+                    }
                 })
                 
                 print(f"{char}: {response}")
-                if context_used:
-                    print(f"  [使用上下文: {', '.join(context_used)}]")
+                # if context_used:
+                #     print(f"  [使用上下文: {', '.join(context_used)}]")
+                
                 print()
         
         return results
     
-def main():
-    """主函数 - 演示系统使用"""
-    print("=== RAG增强双AI角色对话系统 ===")
+    def run_dual_model_conversation(self, session_id: str, topic: str = "", turns: int = 4,
+                                   history_context_count: int = 3, context_info_count: int = 2):
+        """使用双模型系统运行对话
         
-    # 设置路径
-    knowledge_dir = "./knowledge_base"  # 包含世界观和角色文档的目录
+        Args:
+            session_id: 会话ID
+            topic: 对话主题
+            turns: 对话轮数
+            history_context_count: 使用的历史对话轮数
+            context_info_count: 使用的上下文信息数量
+        """
+        # 检查是否为双模型对话系统
+        if not hasattr(self.llm_generator, 'run_dual_character_conversation'):
+            print("⚠ 当前系统不支持双模型对话")
+            return self.run_conversation_turn(session_id, self.llm_generator.list_characters(), turns, topic, 
+                                           history_context_count, context_info_count)
         
-    # 检查必要文件
-    required_dirs = [knowledge_dir]
-    for dir_path in required_dirs:
-        if not os.path.exists(dir_path):
-            print(f"✗ 目录不存在: {dir_path}")
-            print("请确保以下文件存在：")
-            print("- ./knowledge_base/worldview_template_coc.json")
-            print("- ./knowledge_base/character_template_detective.json") 
-            print("- ./knowledge_base/character_template_professor.json")
-            return
+        # 获取对话历史
+        dialogue_history = self.conv_mgr.get_conversation_history(session_id)
         
-    try:
-        # 初始化系统组件
-        print("\n初始化系统...")
-        kb = RAGKnowledgeBase(knowledge_dir)
-        conv_mgr = ConversationManager()
+        # 构建上下文信息
+        if dialogue_history:
+            recent_turns = dialogue_history[-history_context_count:] if history_context_count > 0 else []
+            recent_content = " ".join([turn.content for turn in recent_turns])
+            search_query = recent_content + " " + topic
+        else:
+            search_query = f"{topic} introduction greeting"
         
-        # 这里需要你的LLM生成器，使用现有的NPCDialogueGenerator
-        from npc_dialogue_generator import NPCDialogueGenerator
-        base_model_path = '/mnt/g/Project02/AITrain/Qwen/Qwen3-4B'  # 根据你的路径调整
-        lora_model_path = './output/NPC_Dialogue_LoRA/final_model'
+        # 搜索相关上下文
+        context_info = self.kb.search_relevant_context(search_query, top_k=context_info_count)
         
-        if not os.path.exists(lora_model_path):
-            lora_model_path = None
+        # 构建上下文字符串
+        context_str = ""
+        if context_info:
+            context_str = "相关背景信息："
+            for info in context_info[:context_info_count]:
+                content = info['content'][:150] + "..." if len(info['content']) > 150 else info['content']
+                context_str += f"\n- {content}"
         
-        llm_generator = NPCDialogueGenerator(base_model_path, lora_model_path)
+        print(f"\n=== 双模型对话系统 ===")
+        print(f"主题: {topic}")
+        print(f"角色: {', '.join(self.llm_generator.list_characters())}")
+        print(f"轮数: {turns}")
+        print(f"上下文设置: 历史{history_context_count}轮, 信息{context_info_count}个")
         
-        # 创建对话引擎
-        dialogue_engine = DualAIDialogueEngine(kb, conv_mgr, llm_generator)
+        # 使用双模型系统生成对话
+        for turn in range(turns):
+            # 获取对话历史
+            dialogue_history = self.conv_mgr.get_conversation_history(session_id)
+            conversation_results = self.llm_generator.run_dual_character_conversation(
+                topic=topic,
+                turn_index = turn,
+                context=context_str,
+                dialogue_history = dialogue_history,
+                history_context_count = history_context_count,
+                max_new_tokens=150
+            )
     
-        print("✓ 系统初始化完成")
+            # 保存对话到数据库
+            for result in conversation_results:
+                self.conv_mgr.add_dialogue_turn(
+                    session_id, 
+                    result['speaker'], 
+                    result['dialogue'], 
+                    [result.get('context_used', '')], 
+                    0.8  # 默认相关性分数
+                )
         
-        # 交互式菜单
-        while True:
-            print("\n" + "="*50)
-            print("双AI角色对话系统")
-            print("1. 创建新对话")
-            print("2. 继续已有对话") 
-            print("3. 查看对话历史")
-            print("4. 列出所有会话")
-            print("0. 退出")
-            print("="*50)
 
-            choice = input("请选择操作: ").strip()
+        return conversation_results
 
-            if choice == '0':
-                break
+# def main():
+#     """主函数 - 演示系统使用"""
+#     print("=== RAG增强双AI角色对话系统 ===")
     
-            elif choice == '1':
-                # 创建新对话
-                print(f"可用角色: {list(kb.character_data.keys())}")
-                characters = input("请输入两个角色名称（用空格分隔）: ").strip().split()
+#     # 设置路径
+#     knowledge_dir = "./knowledge_base"  # 包含世界观和角色文档的目录
     
-                if len(characters) != 2:
-                    print("❌ 请输入正好两个角色名称")
-                    continue
+#     # 检查必要文件
+#     required_dirs = [knowledge_dir]
+#     for dir_path in required_dirs:
+#         if not os.path.exists(dir_path):
+#             print(f"✗ 目录不存在: {dir_path}")
+#             print("请确保以下文件存在：")
+#             print("- ./knowledge_base/worldview_template_coc.json")
+#             print("- ./knowledge_base/character_template_detective.json") 
+#             print("- ./knowledge_base/character_template_professor.json")
+#             return
     
-                worldview = kb.worldview_data.get('worldview_name', '未知世界观') if kb.worldview_data else '未知世界观'
-                session_id = conv_mgr.create_session(characters, worldview)
+#     try:
+#         # 初始化系统组件
+#         print("\n初始化系统...")
+#         kb = RAGKnowledgeBase(knowledge_dir)
+#         conv_mgr = ConversationManager()
         
-                topic = input("请输入对话主题（可选）: ").strip()
-                turns = int(input("请输入对话轮次数量（默认2）: ").strip() or "2")
+#         # 这里需要你的LLM生成器，使用新的双模型对话系统
+#         from npc_dialogue_generator import DualModelDialogueGenerator
+#         base_model_path = '/mnt/g/Project02/AITrain/Qwen/Qwen3-4B'  # 根据你的路径调整
+#         lora_model_path = './output/NPC_Dialogue_LoRA/final_model'
         
-                print(f"\n开始对话 - 会话ID: {session_id}")
-                dialogue_engine.run_conversation_turn(session_id, characters, turns, topic)
+#         if not os.path.exists(lora_model_path):
+#             lora_model_path = None
         
-            elif choice == '2':
-                # 继续已有对话
-                sessions = conv_mgr.list_sessions()
-                if not sessions:
-                    print("❌ 没有已有对话")
-                    continue
+#         # 创建双模型对话生成器
+#         if hasattr(kb, 'character_data') and len(kb.character_data) >= 2:
+#             print("✓ 使用knowledge_base角色数据创建双模型对话系统")
+#             # 获取前两个角色
+#             character_names = list(kb.character_data.keys())[:2]
+#             char1_name = character_names[0]
+#             char2_name = character_names[1]
             
-                print("已有会话:")
-                for i, session in enumerate(sessions[:5]):
-                    chars = ", ".join(session['characters'])
-                    print(f"{i+1}. {session['session_id'][:8]}... ({chars}) - {session['last_update'][:16]}")
+#             # 配置两个角色的模型
+#             character1_config = {
+#                 "name": char1_name,
+#                 "lora_path": lora_model_path,  # 可以为每个角色设置不同的LoRA
+#                 "character_data": kb.character_data[char1_name]
+#             }
             
-                try:
-                    idx = int(input("请选择会话编号: ").strip()) - 1
-                    if 0 <= idx < len(sessions):
-                        session = sessions[idx]
-                        session_id = session['session_id']
-                        characters = session['characters']
+#             character2_config = {
+#                 "name": char2_name,
+#                 "lora_path": lora_model_path,  # 可以为每个角色设置不同的LoRA
+#                 "character_data": kb.character_data[char2_name]
+#             }
             
-                        # 显示最近的对话
-                        history = conv_mgr.get_conversation_history(session_id, 4)
-                        if history:
-                            print("\n最近的对话:")
-                            for turn in history:
-                                print(f"{turn.speaker}: {turn.content}")
+#             llm_generator = DualModelDialogueGenerator(
+#                 base_model_path, 
+#                 character1_config, 
+#                 character2_config
+#             )
+#         else:
+#             print("⚠ 角色数据不足，无法创建双模型对话系统")
+#             return
         
-                        topic = input("请输入对话主题（可选）: ").strip()
-                        turns = int(input("请输入对话轮次数量（默认1）: ").strip() or "1")
+#         # 创建对话引擎
+#         dialogue_engine = DualAIDialogueEngine(kb, conv_mgr, llm_generator)
         
-                        print(f"\n继续对话 - 会话ID: {session_id}")
-                        dialogue_engine.run_conversation_turn(session_id, characters, turns, topic)
-                    else:
-                        print("❌ 无效的会话编号")
-                except ValueError:
-                    print("❌ 请输入有效的数字")
+#         print("✓ 系统初始化完成")
         
-            elif choice == '3':
-                # 查看对话历史
-                session_id = input("请输入会话ID（前8位即可）: ").strip()
+#         # 交互式菜单
+#         while True:
+#             print("\n" + "="*50)
+#             print("双AI角色对话系统")
+#             print("1. 创建新对话")
+#             print("2. 继续已有对话") 
+#             print("3. 查看对话历史")
+#             print("4. 列出所有会话")
+#             print("0. 退出")
+#             print("="*50)
             
-                # 查找匹配的会话
-                sessions = conv_mgr.list_sessions()
-                matching_session = None
-                for session in sessions:
-                    if session['session_id'].startswith(session_id):
-                        matching_session = session
-                        break
+#             choice = input("请选择操作: ").strip()
             
-                if matching_session:
-                    full_session_id = matching_session['session_id']
-                    history = conv_mgr.get_conversation_history(full_session_id, 20)
+#             if choice == '0':
+#                 break
                 
-                    if history:
-                        print(f"\n对话历史 - {full_session_id}")
-                        print(f"角色: {', '.join(matching_session['characters'])}")
-                        print(f"世界观: {matching_session['worldview']}")
-                        print("-" * 50)
+#             elif choice == '1':
+#                 # 创建新对话
+#                 print(f"可用角色: {list(kb.character_data.keys())}")
+#                 characters = input("请输入两个角色名称（用空格分隔）: ").strip().split()
                 
-                        for turn in history:
-                            print(f"[{turn.timestamp[:16]}] {turn.speaker}:")
-                            print(f"  {turn.content}")
-                            if turn.context_used:
-                                print(f"  使用上下文: {', '.join(turn.context_used)}")
-                            print()
-                    else:
-                        print("该会话暂无对话历史")
-                else:
-                    print("❌ 未找到匹配的会话")
+#                 if len(characters) != 2:
+#                     print("❌ 请输入正好两个角色名称")
+#                     continue
                 
-            elif choice == '4':
-                # 列出所有会话
-                sessions = conv_mgr.list_sessions()
-                if sessions:
-                    print(f"\n共有 {len(sessions)} 个对话会话:")
-                    for session in sessions:
-                        chars = ", ".join(session['characters'])
-                        print(f"ID: {session['session_id']}")
-                        print(f"  角色: {chars}")
-                        print(f"  世界观: {session['worldview']}")
-                        print(f"  最后更新: {session['last_update']}")
-                        print()
-                else:
-                    print("暂无对话会话")
+#                 worldview = kb.worldview_data.get('worldview_name', '未知世界观') if kb.worldview_data else '未知世界观'
+#                 session_id = conv_mgr.create_session(characters, worldview)
                 
-            else:
-                print("❌ 无效选择")
+#                 topic = input("请输入对话主题（可选）: ").strip()
+#                 turns = int(input("请输入对话轮次数量（默认2）: ").strip() or "2")
                 
-    except Exception as e:
-        print(f"✗ 系统运行出错: {e}")
-        import traceback
-        traceback.print_exc()
+#                 # 历史上下文控制选项
+#                 print("\n历史上下文设置:")
+#                 history_count = input("使用历史对话轮数（默认3，0表示不使用）: ").strip()
+#                 history_count = int(history_count) if history_count.isdigit() else 3
                 
-if __name__ == '__main__':
-    main()
+#                 context_info_count = input("使用上下文信息数量（默认2）: ").strip()
+#                 context_info_count = int(context_info_count) if context_info_count.isdigit() else 2
+                
+#                 print(f"\n开始对话 - 会话ID: {session_id}")
+#                 print(f"上下文设置: 历史{history_count}轮, 信息{context_info_count}个")
+                
+#                 # 询问是否使用双模型对话
+#                 use_dual_model = input("是否使用双模型对话系统？(y/n，默认y): ").strip().lower()
+#                 if use_dual_model != 'n':
+#                     print("使用双模型对话系统...")
+#                     dialogue_engine.run_dual_model_conversation(session_id, topic, turns, history_count, context_info_count)
+#                 else:
+#                     print("使用传统对话系统...")
+#                     dialogue_engine.run_conversation_turn(session_id, characters, turns, topic, history_count, context_info_count)
+                
+#             elif choice == '2':
+#                 # 继续已有对话
+#                 sessions = conv_mgr.list_sessions()
+#                 if not sessions:
+#                     print("❌ 没有已有对话")
+#                     continue
+                
+#                 print("已有会话:")
+#                 for i, session in enumerate(sessions[:5]):
+#                     chars = ", ".join(session['characters'])
+#                     print(f"{i+1}. {session['session_id'][:8]}... ({chars}) - {session['last_update'][:16]}")
+                
+#                 try:
+#                     idx = int(input("请选择会话编号: ").strip()) - 1
+#                     if 0 <= idx < len(sessions):
+#                         session = sessions[idx]
+#                         session_id = session['session_id']
+#                         characters = session['characters']
+                        
+#                         # 显示最近的对话
+#                         history = conv_mgr.get_conversation_history(session_id, 4)
+#                         if history:
+#                             print("\n最近的对话:")
+#                             for turn in history:
+#                                 print(f"{turn.speaker}: {turn.content}")
+                        
+#                         topic = input("请输入对话主题（可选）: ").strip()
+#                         turns = int(input("请输入对话轮次数量（默认1）: ").strip() or "1")
+                        
+#                         # 历史上下文控制选项
+#                         print("\n历史上下文设置:")
+#                         history_count = input("使用历史对话轮数（默认3，0表示不使用）: ").strip()
+#                         history_count = int(history_count) if history_count.isdigit() else 3
+                        
+#                         context_info_count = input("使用上下文信息数量（默认2）: ").strip()
+#                         context_info_count = int(context_info_count) if context_info_count.isdigit() else 2
+                        
+#                         print(f"\n继续对话 - 会话ID: {session_id}")
+#                         print(f"上下文设置: 历史{history_count}轮, 信息{context_info_count}个")
+                        
+#                         # 询问是否使用双模型对话
+#                         use_dual_model = input("是否使用双模型对话系统？(y/n，默认y): ").strip().lower()
+#                         if use_dual_model != 'n':
+#                             print("使用双模型对话系统...")
+#                             dialogue_engine.run_dual_model_conversation(session_id, topic, turns, history_count, context_info_count)
+#                         else:
+#                             print("使用传统对话系统...")
+#                             dialogue_engine.run_conversation_turn(session_id, characters, turns, topic, history_count, context_info_count)
+#                     else:
+#                         print("❌ 无效的会话编号")
+#                 except ValueError:
+#                     print("❌ 请输入有效的数字")
+                    
+#             elif choice == '3':
+#                 # 查看对话历史
+#                 session_id = input("请输入会话ID（前8位即可）: ").strip()
+                
+#                 # 查找匹配的会话
+#                 sessions = conv_mgr.list_sessions()
+#                 matching_session = None
+#                 for session in sessions:
+#                     if session['session_id'].startswith(session_id):
+#                         matching_session = session
+#                         break
+                
+#                 if matching_session:
+#                     full_session_id = matching_session['session_id']
+#                     history = conv_mgr.get_conversation_history(full_session_id, 20)
+                    
+#                     if history:
+#                         print(f"\n对话历史 - {full_session_id}")
+#                         print(f"角色: {', '.join(matching_session['characters'])}")
+#                         print(f"世界观: {matching_session['worldview']}")
+#                         print("-" * 50)
+                        
+#                         for turn in history:
+#                             print(f"[{turn.timestamp[:16]}] {turn.speaker}:")
+#                             print(f"  {turn.content}")
+#                             if turn.context_used:
+#                                 print(f"  使用上下文: {', '.join(turn.context_used)}")
+#                             print()
+#                     else:
+#                         print("该会话暂无对话历史")
+#                 else:
+#                     print("❌ 未找到匹配的会话")
+                    
+#             elif choice == '4':
+#                 # 列出所有会话
+#                 sessions = conv_mgr.list_sessions()
+#                 if sessions:
+#                     print(f"\n共有 {len(sessions)} 个对话会话:")
+#                     for session in sessions:
+#                         chars = ", ".join(session['characters'])
+#                         print(f"ID: {session['session_id']}")
+#                         print(f"  角色: {chars}")
+#                         print(f"  世界观: {session['worldview']}")
+#                         print(f"  最后更新: {session['last_update']}")
+#                         print()
+#                 else:
+#                     print("暂无对话会话")
+            
+#             else:
+#                 print("❌ 无效选择")
+    
+#     except Exception as e:
+#         print(f"✗ 系统运行出错: {e}")
+#         import traceback
+#         traceback.print_exc()
+
+# if __name__ == '__main__':
+#     main()

AITrain/knowledge_base/character_template_detective.json

@@ -0,0 +1,182 @@
+{
+  "character_name": "维多利亚·布莱克伍德",
+  "basic_info": {
+    "age": 28,
+    "gender": "女",
+    "occupation": "私人侦探",
+    "nationality": "英国",
+    "appearance": "身材高挑，有着深邃的绿色眼睛和波浪状的棕色长发，通常穿着深色长外套"
+  },
+  "personality": {
+    "core_traits": [
+      "理性冷静",
+      "观察力敏锐",
+      "独立自主",
+      "好奇心强"
+    ],
+    "strengths": [
+      "逻辑推理能力强",
+      "善于发现细节",
+      "不畏危险",
+      "坚持正义"
+    ],
+    "weaknesses": [
+      "有时过于固执",
+      "不善表达情感",
+      "容易忽视他人感受",
+      "工作狂倾向"
+    ],
+    "values": [
+      "真相高于一切",
+      "保护无辜的人",
+      "相信科学与理性",
+      "追求公正"
+    ]
+  },
+  "background": {
+    "childhood": "出生在伦敦一个中产阶级家庭，父亲是律师，母亲是医生。从小就表现出对解谜的兴趣。",
+    "education": "剑桥大学法学专业毕业，后来转向犯罪学研究",
+    "career_start": "最初在苏格兰场工作了3年，后来决定成为私人侦探",
+    "major_events": [
+      "21岁时目睹了一起谋杀案，激发了对侦探工作的热情",
+      "25岁时成功破获了一起连环杀手案，声名鹊起",
+      "27岁时开设了自己的私人侦探事务所"
+    ]
+  },
+  "skills_and_abilities": {
+    "professional_skills": [
+      "犯罪现场调查",
+      "法医学基础",
+      "心理分析",
+      "跟踪监视",
+      "档案研究"
+    ],
+    "languages": [
+      "英语(母语)",
+      "法语(流利)",
+      "德语(基础)"
+    ],
+    "other_abilities": [
+      "摄影",
+      "速记",
+      "基础格斗技能",
+      "开锁技术"
+    ]
+  },
+  "relationships": {
+    "allies": [
+      {
+        "name": "詹姆斯·沃森医生",
+        "relationship": "合作伙伴",
+        "description": "经常协助处理涉及医学知识的案件"
+      },
+      {
+        "name": "艾米丽·哈里斯警官",
+        "relationship": "朋友/信息源",
+        "description": "苏格兰场的老同事，偶尔提供官方信息"
+      }
+    ],
+    "family": [
+      {
+        "name": "威廉·布莱克伍德",
+        "relationship": "父亲",
+        "description": "资深律师，关系良好但不常联系"
+      },
+      {
+        "name": "玛格丽特·布莱克伍德",
+        "relationship": "母亲",
+        "description": "内科医生，担心女儿的安危但支持她的选择"
+      }
+    ]
+  },
+  "speech_patterns": {
+    "vocabulary": [
+      "倾向使用准确的专业词汇",
+      "很少使用情绪化的表达",
+      "喜欢引用事实和数据",
+      "经常使用推理性语句"
+    ],
+    "tone": [
+      "语调平稳",
+      "说话简洁明了",
+      "很少有语气词",
+      "在紧张时语速会变快"
+    ],
+    "habits": [
+      "习惯在思考时轻敲桌面",
+      "总是随身携带笔记本",
+      "喜欢用问句引导对话",
+      "经常说'让我们看看证据'"
+    ],
+    "sample_phrases": [
+      "根据现有证据...",
+      "这很有趣...",
+      "让我们从逻辑角度分析",
+      "事实胜于推测",
+      "细节决定成败"
+    ]
+  },
+  "current_situation": {
+    "location": "伦敦贝克街221B的侦探事务所",
+    "recent_activities": [
+      "刚刚结束了一起艺术品盗窃案的调查",
+      "正在研究一系列看似无关的失踪案件",
+      "计划更新事务所的调查设备"
+    ],
+    "current_mood": "保持专业警觉，对新案件充满期待",
+    "goals": [
+      "短期：解决手头的失踪案件",
+      "中期：扩大事务所的影响力",
+      "长期：成为伦敦最知名的私人侦探"
+    ]
+  },
+  "preferences": {
+    "likes": [
+      "复杂的推理谜题",
+      "古典音乐",
+      "黑咖啡",
+      "下雨天的伦敦",
+      "老式侦探小说"
+    ],
+    "dislikes": [
+      "无意义的社交",
+      "被人轻视",
+      "不完整的信息",
+      "时间被浪费",
+      "感情用事的决定"
+    ],
+    "habits": [
+      "每天早上6点起床",
+      "喜欢在深夜整理案件资料",
+      "周末会去博物馆或图书馆",
+      "睡前必须检查门锁"
+    ]
+  },
+  "dialogue_examples": {
+    "casual_conversation": [
+      "我觉得这个案子比表面看起来更复杂。",
+      "证据链还不完整，我们需要更多信息。",
+      "有趣，这个细节之前被忽略了。"
+    ],
+    "professional_interaction": [
+      "请详细描述当时的情况，不要遗漏任何细节。",
+      "根据我的调查，事情的真相可能是这样的...",
+      "我需要检查现场，任何痕迹都可能是关键。"
+    ],
+    "emotional_moments": [
+      "有时候，真相比谎言更难以接受。",
+      "正义可能会迟到，但绝不会缺席。",
+      "每个案件背后都有真实的人生。"
+    ]
+  },
+  "meta_info": {
+    "creation_purpose": "用于COC跑团或现代侦探类角色扮演",
+    "adaptability": "可根据具体剧情需要调整背景和关系网",
+    "compatibility": "适合与其他侦探、医生、警察等角色配合",
+    "character_arc_potential": [
+      "从冷静理性到学会关心他人情感",
+      "从独立工作到学会团队协作",
+      "面对更大阴谋时的成长"
+    ]
+  }
+}

AITrain/knowledge_base/character_template_professor.json

@@ -0,0 +1,242 @@
+{
+  "character_name": "阿奇博尔德·韦恩",
+  "basic_info": {
+    "age": 45,
+    "gender": "男",
+    "occupation": "米斯卡托尼克大学考古学教授",
+    "nationality": "美国",
+    "appearance": "中等身材，银白色头发，总是戴着金丝边眼镜，穿着考究的三件套西装"
+  },
+  "personality": {
+    "core_traits": [
+      "博学严谨",
+      "谨慎保守",
+      "责任心强",
+      "内心焦虑"
+    ],
+    "strengths": [
+      "渊博的学识",
+      "精确的记忆力",
+      "出色的研究能力",
+      "对学生的关心"
+    ],
+    "weaknesses": [
+      "过度担心",
+      "决策犹豫",
+      "身体较弱",
+      "容易陷入研究忘记现实"
+    ],
+    "values": [
+      "知识的传承",
+      "学术诚信",
+      "保护无知者",
+      "维护理性"
+    ]
+  },
+  "background": {
+    "childhood": "出生在波士顿一个书香门第，父亲是历史学教授，从小在图书馆中长大",
+    "education": "哈佛大学考古学博士，专精古代文明研究，特别是前哥伦布时期美洲文明",
+    "career_development": "29岁进入米斯卡托尼克大学任教，逐渐接触到一些'不寻常'的考古发现",
+    "major_events": [
+      "35岁时参与了一次南美洲的考古挖掘，发现了一些令人不安的古老雕像",
+      "40岁时首次接触到《死灵之书》的片段，从此开始关注超自然考古学",
+      "42岁时目睹了一次超自然事件，开始理解世界的真实面貌",
+      "目前正在秘密研究如何保护人类免受古老威胁"
+    ]
+  },
+  "skills_and_abilities": {
+    "academic_skills": [
+      "古代语言学（拉丁语、古希腊语、古玛雅文字）",
+      "考古学田野调查",
+      "古代文明研究",
+      "文物鉴定与保护",
+      "学术论文写作"
+    ],
+    "occult_knowledge": [
+      "神秘学基础理论",
+      "古老符号的辨识",
+      "保护性仪式",
+      "危险文物的识别",
+      "理智保护技巧"
+    ],
+    "practical_abilities": [
+      "图书馆研究",
+      "多种古代语言翻译",
+      "文献考证",
+      "教学能力",
+      "基础摄影技术"
+    ]
+  },
+  "relationships": {
+    "academic_circles": [
+      {
+        "name": "亨利·阿米蒂奇教授",
+        "relationship": "同事兼导师",
+        "description": "图书馆学教授，是少数知道真相的学者之一"
+      },
+      {
+        "name": "弗朗西斯·摩根教授",
+        "relationship": "研究伙伴",
+        "description": "医学教授，专门研究超自然现象对人体的影响"
+      }
+    ],
+    "students": [
+      {
+        "name": "多名研究生",
+        "relationship": "师生",
+        "description": "尽力保护他们不接触危险知识，同时培养他们的学术能力"
+      }
+    ],
+    "family": [
+      {
+        "name": "伊莲娜·韦恩",
+        "relationship": "妻子",
+        "description": "中学英语教师，不知道丈夫研究的真实性质，经常担心他的健康"
+      },
+      {
+        "name": "托马斯·韦恩",
+        "relationship": "儿子",
+        "description": "16岁，对父亲的工作很好奇，韦恩博士努力让他远离危险知识"
+      }
+    ]
+  },
+  "speech_patterns": {
+    "vocabulary": [
+      "经常使用学术术语",
+      "引用古典文献",
+      "措辞谨慎保守",
+      "避免直接描述恐怖事物"
+    ],
+    "tone": [
+      "语调温和但认真",
+      "说话时经常停顿思考",
+      "在谈到危险知识时声音会变低",
+      "对学生总是耐心解释"
+    ],
+    "habits": [
+      "经常擦拭眼镜",
+      "思考时会无意识地整理桌上的文件",
+      "喜欢用'据我所知'开头",
+      "经常提及'需要更多研究'"
+    ],
+    "sample_phrases": [
+      "根据我的研究...",
+      "这需要极其谨慎的处理",
+      "古人的智慧不容小觑",
+      "有些知识是有代价的",
+      "我们必须保护那些不知情的人"
+    ]
+  },
+  "current_situation": {
+    "location": "米斯卡托尼克大学考古系办公室",
+    "recent_activities": [
+      "正在翻译一份神秘的古代文献",
+      "与其他几位教授秘密讨论如何应对潜在威胁",
+      "准备下个月的美洲古文明课程",
+      "关注最近发生的几起奇异事件"
+    ],
+    "current_concerns": [
+      "担心某个学生可能接触到了危险知识",
+      "对最近校园里的奇怪现象感到不安",
+      "妻子开始怀疑他隐瞒什么事情"
+    ],
+    "goals": [
+      "短期：完成当前文献的翻译工作",
+      "中期：建立更有效的超自然威胁预警系统",
+      "长期：培养能够承担保护人类责任的下一代学者"
+    ]
+  },
+  "mental_state": {
+    "sanity_level": "相对稳定，但承受着巨大心理压力",
+    "coping_mechanisms": [
+      "通过规律的学术工作保持理性",
+      "与信任的同事分享担忧",
+      "严格限制自己接触最危险的知识",
+      "通过教学工作获得成就感"
+    ],
+    "stress_indicators": [
+      "失眠和噩梦",
+      "过度担心家人安全",
+      "对学生过分保护",
+      "经常检查办公室的安全性"
+    ]
+  },
+  "preferences": {
+    "likes": [
+      "古典音乐，特别是巴赫",
+      "安静的图书馆环境",
+      "与同行讨论安全的学术话题",
+      "和家人共度的平静时光",
+      "收集古代艺术品（安全的那些）"
+    ],
+    "dislikes": [
+      "不负责任的学者",
+      "对超自然现象的轻率态度",
+      "媒体对考古发现的夸大报道",
+      "任何可能危及学生的情况",
+      "不得不对家人撒谎"
+    ],
+    "daily_habits": [
+      "早上6点起床，先查看昨夜是否有异常",
+      "上午专注于教学工作",
+      "午后在办公室进行研究",
+      "晚上陪伴家人，避免谈论工作",
+      "睡前检查所有门窗"
+    ]
+  },
+  "equipment_and_possessions": {
+    "academic_tools": [
+      "高质量的放大镜和测量工具",
+      "多种古代语言词典",
+      "精密的摄影设备",
+      "文物保护用品",
+      "大量专业书籍"
+    ],
+    "protective_items": [
+      "一些他认为有保护作用的符咒",
+      "经过特殊处理的盐",
+      "几本关于保护仪式的手册",
+      "应急通讯设备"
+    ],
+    "personal_belongings": [
+      "妻子的照片",
+      "父亲留下的金表",
+      "学生送的纪念品",
+      "一些'安全'的古代复制品"
+    ]
+  },
+  "dialogue_examples": {
+    "teaching_moments": [
+      "古代文明留给我们的不仅仅是石头和黄金，更重要的是智慧...和警告。",
+      "在考古学中，我们不仅要发现过去，更要保护现在。",
+      "有些发现最好留在地下，这不是怯懦，而是智慧。"
+    ],
+    "research_discussions": [
+      "这些符号...我在其他地方见过，但愿我的猜测是错的。",
+      "我们需要更多的资料，但要确保安全地获取。",
+      "如果我的翻译是正确的，那么我们面临的威胁比想象中更严重。"
+    ],
+    "personal_conversations": [
+      "对不起，亲爱的，最近工作压力确实很大。",
+      "托马斯，也许你应该考虑学习更实用的专业，比如工程学？",
+      "有些时候，无知确实是一种幸福。"
+    ],
+    "crisis_responses": [
+      "保持冷静，记住我们的训练。",
+      "不要直视它！专注于我们讨论过的保护咒语。",
+      "如果我们失败了，确保销毁所有相关文献。"
+    ]
+  },
+  "character_arc_potential": [
+    "从纯粹的学者成长为保护人类的守护者",
+    "在家庭责任和拯救世界之间寻找平衡",
+    "面对更大威胁时必须做出牺牲的决定",
+    "培养继承者以延续保护人类的使命"
+  ],
+  "meta_info": {
+    "creation_purpose": "COC跑团中的学者型NPC，可以提供知识支援和道德指导",
+    "role_in_story": "导师、信息来源、情感支柱",
+    "interaction_style": "谨慎但关怀，知识丰富但不会直接给出所有答案",
+    "narrative_function": "平衡角色团队，提供背景知识，增加故事深度"
+  }
+}

AITrain/knowledge_base/worldview_template_coc.json

@@ -0,0 +1,296 @@
+{
+  "worldview_name": "克苏鲁的呼唤 (Call of Cthulhu)",
+  "basic_info": {
+    "genre": "恐怖，超自然，推理",
+    "time_period": "1920年代",
+    "primary_setting": "美国新英格兰地区",
+    "tone": "黑暗，压抑，充满未知恐惧"
+  },
+  "core_concepts": {
+    "cosmic_horror": {
+      "description": "人类在宇宙中的渺小和无知",
+      "key_elements": [
+        "古老存在远超人类理解",
+        "知识本身就是危险的",
+        "人类理性的脆弱性",
+        "宇宙的冷漠和敌意"
+      ]
+    },
+    "sanity_system": {
+      "description": "理智值系统，接触超自然会损害心智",
+      "mechanics": [
+        "目睹恐怖事物降低理智",
+        "学习禁忌知识的代价",
+        "疯狂的不同表现形式",
+        "理智恢复的困难性"
+      ]
+    },
+    "mythos_knowledge": {
+      "description": "关于古老神话的危险知识",
+      "categories": [
+        "旧日支配者",
+        "外神",
+        "禁忌典籍",
+        "异界生物"
+      ]
+    }
+  },
+  "geography": {
+    "primary_locations": [
+      {
+        "name": "阿卡姆",
+        "description": "马萨诸塞州的大学城，米斯卡托尼克大学所在地",
+        "significance": "学术研究中心，许多超自然事件的发源地",
+        "notable_features": [
+          "米斯卡托尼克大学图书馆",
+          "阿卡姆疗养院",
+          "各种古老建筑"
+        ]
+      },
+      {
+        "name": "敦威治",
+        "description": "偏远的乡村小镇，充满古老传说",
+        "significance": "许多怪异事件的发生地",
+        "notable_features": [
+          "古老的山丘",
+          "废弃的农场",
+          "神秘的石圈"
+        ]
+      },
+      {
+        "name": "印斯茅斯",
+        "description": "衰败的海港小镇",
+        "significance": "深潜者的据点",
+        "notable_features": [
+          "古老的码头",
+          "奇怪的居民",
+          "海底的秘密"
+        ]
+      }
+    ],
+    "other_significant_places": [
+      "南极洲的疯狂山脉",
+      "太平洋的拉莱耶",
+      "梦境大陆",
+      "时空的裂隙点"
+    ]
+  },
+  "timeline": {
+    "ancient_times": [
+      "数百万年前：旧日支配者统治地球",
+      "人类出现前：古老种族的兴衰",
+      "史前时代：各种异界生物的活动"
+    ],
+    "historical_events": [
+      "1692年：塞勒姆女巫审判（隐藏的真实原因）",
+      "1846年：某次南极探险的神秘失踪",
+      "1908年：通古斯大爆炸的真相",
+      "1912年：泰坦尼克号沉没的隐情"
+    ],
+    "current_era_1920s": [
+      "1920年：禁酒令开始，地下活动增加",
+      "1922年：某次考古发现引发的事件",
+      "1925年：格陵兰探险队的失踪",
+      "1929年：股市崩盘前的预兆事件"
+    ]
+  },
+  "factions_and_organizations": {
+    "academic": [
+      {
+        "name": "米斯卡托尼克大学",
+        "purpose": "高等教育和研究机构",
+        "secret_role": "神秘学知识的守护者",
+        "notable_members": [
+          "亨利·阿米蒂奇教授",
+          "其他不愿透露姓名的学者"
+        ]
+      }
+    ],
+    "occult": [
+      {
+        "name": "星之智慧教派",
+        "purpose": "崇拜外神和旧日支配者",
+        "activities": [
+          "进行邪恶仪式",
+          "寻找古老遗物",
+          "传播危险知识"
+        ]
+      }
+    ],
+    "government": [
+      {
+        "name": "联邦调查局（非正式部门）",
+        "purpose": "调查超自然事件",
+        "secrecy_level": "高度机密",
+        "resources": "有限但专业"
+      }
+    ]
+  },
+  "supernatural_elements": {
+    "entities": {
+      "old_ones": [
+        {
+          "name": "克苏鲁",
+          "domain": "海洋，梦境",
+          "status": "沉睡中",
+          "influence": "通过梦境影响敏感者"
+        },
+        {
+          "name": "哈斯塔",
+          "domain": "风，空气，艺术",
+          "symbols": "黄衣之王",
+          "influence": "通过艺术作品传播疯狂"
+        }
+      ],
+      "lesser_beings": [
+        "深潜者",
+        "食尸鬼",
+        "夜魇",
+        "拜亚基"
+      ]
+    },
+    "artifacts": [
+      {
+        "name": "死灵之书（Necronomicon）",
+        "type": "禁忌典籍",
+        "danger_level": "极高",
+        "effects": "提供强大知识但严重损害理智"
+      },
+      {
+        "name": "闪亮的偏方三八面体",
+        "type": "几何物品",
+        "properties": "可以打开时空裂隙"
+      }
+    ]
+  },
+  "social_context": {
+    "1920s_america": {
+      "culture": [
+        "爵士时代的繁荣",
+        "禁酒令下的地下文化",
+        "科学理性主义的兴起",
+        "传统价值观的变化"
+      ],
+      "technology": [
+        "无线电的普及",
+        "汽车开始普及",
+        "电话系统发展",
+        "早期航空业"
+      ],
+      "social_issues": [
+        "种族隔离",
+        "性别不平等",
+        "经济不稳定",
+        "城市化进程"
+      ]
+    },
+    "academic_world": [
+      "大学的黄金时代",
+      "考古学的发展",
+      "人类学研究的兴起",
+      "心理学的新发现"
+    ]
+  },
+  "themes": {
+    "primary_themes": [
+      "人类在宇宙中的渺小",
+      "知识的双刃剑性质",
+      "文明的脆弱性",
+      "理性与疯狂的边界"
+    ],
+    "common_scenarios": [
+      "调查神秘失踪案件",
+      "发现古老遗迹",
+      "面对无法解释的现象",
+      "与邪教组织斗争",
+      "保护无知的大众"
+    ],
+    "moral_dilemmas": [
+      "是否应该追求危险的真相",
+      "如何平衡知识与安全",
+      "是否要向他人透露恐怖真相",
+      "如何在保持理智的同时对抗超自然"
+    ]
+  },
+  "gameplay_elements": {
+    "investigation": [
+      "收集线索",
+      "访问证人",
+      "研究历史档案",
+      "解读古老文本"
+    ],
+    "survival": [
+      "保持理智值",
+      "避免直接接触超自然存在",
+      "寻找盟友和资源",
+      "制定逃脱计划"
+    ],
+    "character_development": [
+      "技能的提升",
+      "知识的积累",
+      "心理创伤的处理",
+      "人际关系的建立"
+    ]
+  },
+  "atmosphere": {
+    "visual_elements": [
+      "雾气弥漫的街道",
+      "古老而阴暗的建筑",
+      "昏暗的图书馆和博物馆",
+      "荒凉的海岸线"
+    ],
+    "audio_elements": [
+      "远处传来的奇怪声音",
+      "老式留声机的音乐",
+      "风吹过古老建筑的声音",
+      "夜晚的虫鸣和海浪声"
+    ],
+    "emotional_tone": [
+      "不安和焦虑",
+      "对未知的恐惧",
+      "孤独和绝望",
+      "偶尔的希望之光"
+    ]
+  },
+  "language_and_style": {
+    "writing_style": [
+      "维多利亚时期的正式语言",
+      "学术性的措辞",
+      "大量使用形容词",
+      "暗示而非直接描述恐怖"
+    ],
+    "common_vocabulary": [
+      "古老的 (ancient)",
+      "不可名状的 (unspeakable)",
+      "亵渎的 (blasphemous)",
+      "令人恐惧的 (dreadful)",
+      "神秘的 (mysterious)"
+    ],
+    "character_speech": [
+      "教授：学术性，引用经典",
+      "调查员：实用主义，关注细节",
+      "普通民众：迷信，害怕未知",
+      "邪教徒：狂热，晦涩难懂"
+    ]
+  },
+  "usage_guidelines": {
+    "for_roleplay": [
+      "强调角色的脆弱性",
+      "重视调查过程",
+      "营造紧张氛围",
+      "合理使用恐怖元素"
+    ],
+    "for_storytelling": [
+      "逐渐揭示真相",
+      "保持神秘感",
+      "平衡恐怖与希望",
+      "关注角色成长"
+    ],
+    "common_pitfalls": [
+      "避免过度描述恐怖",
+      "不要让玩家过于无力",
+      "保持世界的一致性",
+      "尊重原著精神"
+    ]
+  }
+}

AITrain/main_controller.py

@@ -0,0 +1,456 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+'''
+双AI角色对话系统主控制程序
+完整的工作流程：PDF处理 -> 角色加载 -> RAG对话 -> 历史记录
+'''
+
+import os
+import sys
+import shutil
+from typing import List, Dict
+import json
+
+def check_dependencies():
+    """检查依赖库"""
+    missing_deps = []
+    
+    try:
+        import PyPDF2
+    except ImportError:
+        missing_deps.append("PyPDF2")
+    
+    try:
+        import pymupdf
+        print("✓ pymupdf 可用")
+    except ImportError:
+        print("⚠ pymupdf 不可用，将使用 PyPDF2")
+    
+    try:
+        import sentence_transformers
+        import faiss
+        print("✓ 向量化功能可用")
+    except ImportError:
+        print("⚠ 向量化功能不可用，将使用文本匹配")
+    
+    if missing_deps:
+        print(f"✗ 缺少依赖库: {', '.join(missing_deps)}")
+        print("请运行: pip install PyPDF2 sentence-transformers faiss-cpu")
+        return False
+    
+    return True
+
+def setup_directories():
+    """设置项目目录结构"""
+    directories = [
+        "./knowledge_base",
+        "./characters", 
+        "./worldview",
+        "./rag_knowledge",
+        "./conversation_data"
+    ]
+    
+    for dir_path in directories:
+        os.makedirs(dir_path, exist_ok=True)
+        print(f"✓ 目录就绪: {dir_path}")
+
+def copy_demo_files():
+    """复制演示文档到知识库目录"""
+    file_mappings = [
+        ("./worldview/worldview_template_coc.json", "./knowledge_base/worldview_template_coc.json"),
+        ("./characters/character_template_detective.json", "./knowledge_base/character_template_detective.json"),
+        ("./characters/character_template_professor.json", "./knowledge_base/character_template_professor.json")
+    ]
+    
+    for source, target in file_mappings:
+        if os.path.exists(source):
+            shutil.copy2(source, target)
+            print(f"✓ 复制文档: {os.path.basename(target)}")
+
+def process_pdf_workflow():
+    """PDF处理工作流"""
+    print("\n" + "="*60)
+    print("PDF世界观文档处理")
+    print("="*60)
+    
+    from pdf_to_rag_processor import PDFToRAGProcessor
+    
+    pdf_path = input("请输入PDF文件路径 (例: ./coc.pdf): ").strip()
+    
+    if not os.path.exists(pdf_path):
+        print(f"✗ 文件不存在: {pdf_path}")
+        return False
+    
+    try:
+        processor = PDFToRAGProcessor()
+        result = processor.process_pdf_to_rag(pdf_path, "./rag_knowledge")
+        
+        print(f"\n✓ PDF处理完成!")
+        print(f"  - 文档块数: {result['chunks_count']}")
+        print(f"  - 概念数: {result['concepts_count']}")
+        print(f"  - 向量索引: {'启用' if result['vector_enabled'] else '未启用'}")
+        
+        return True
+        
+    except Exception as e:
+        print(f"✗ PDF处理失败: {e}")
+        return False
+
+def show_character_info():
+    """显示角色信息"""
+    print("\n" + "="*60)
+    print("角色设定信息")
+    print("="*60)
+    
+    knowledge_dir = "./knowledge_base"
+    character_files = [f for f in os.listdir(knowledge_dir) if f.startswith('character') and f.endswith('.json')]
+    
+    for char_file in character_files:
+        try:
+            with open(os.path.join(knowledge_dir, char_file), 'r', encoding='utf-8') as f:
+                char_data = json.load(f)
+            
+            name = char_data.get('character_name', '未知')
+            occupation = char_data.get('basic_info', {}).get('occupation', '未知')
+            traits = char_data.get('personality', {}).get('core_traits', [])
+            
+            print(f"\n角色: {name}")
+            print(f"  职业: {occupation}")
+            print(f"  特点: {', '.join(traits[:3])}")
+            
+        except Exception as e:
+            print(f"✗ 读取角色文件失败: {char_file} - {e}")
+
+def run_dialogue_system():
+    """运行双AI对话系统"""
+    print("\n" + "="*60)
+    print("启动双AI角色对话系统")
+    print("="*60)
+    
+    try:
+       
+        # 直接启动双模型对话
+        print("\n正在初始化双模型对话系统...")
+        
+        from dual_ai_dialogue_system import RAGKnowledgeBase, ConversationManager, DualAIDialogueEngine
+        from npc_dialogue_generator import DualModelDialogueGenerator
+        
+        # 初始化组件
+        kb = RAGKnowledgeBase("./knowledge_base")
+        conv_mgr = ConversationManager("./conversation_data/conversations.db")
+        
+        # 检查模型路径
+        base_model_path = '/mnt/e/AI/Project02/AITrain/Qwen/Qwen3-4B'
+        lora_model_path = './output/NPC_Dialogue_LoRA/final_model'
+        
+        if not os.path.exists(base_model_path):
+            print(f"✗ 基础模型路径不存在: {base_model_path}")
+            print("请修改 main_controller.py 中的模型路径")
+            return
+        
+        if not os.path.exists(lora_model_path):
+            lora_model_path = None
+            print("⚠ LoRA模型不存在，使用基础模型")
+        
+        # 检查角色数据
+        if not hasattr(kb, 'character_data') or len(kb.character_data) < 2:
+            print("✗ 角色数据不足，无法创建双模型对话系统")
+            print("请确保knowledge_base目录中有至少两个角色文件")
+            return
+        
+        # 获取前两个角色
+        character_names = list(kb.character_data.keys())[:2]
+        char1_name = character_names[0]
+        char2_name = character_names[1]
+        
+        print(f"✓ 使用角色: {char1_name} 和 {char2_name}")
+        
+        # 配置两个角色的模型
+        character1_config = {
+            "name": char1_name,
+            "lora_path": lora_model_path,
+            "character_data": kb.character_data[char1_name]
+        }
+        
+        character2_config = {
+            "name": char2_name,
+            "lora_path": lora_model_path,
+            "character_data": kb.character_data[char2_name]
+        }
+        
+        # 创建双模型对话生成器
+        print("正在初始化双模型对话生成器...")
+        dual_generator = DualModelDialogueGenerator(
+            base_model_path, 
+            character1_config, 
+            character2_config
+        )
+        
+        # 创建对话引擎
+        dialogue_engine = DualAIDialogueEngine(kb, conv_mgr, dual_generator)
+        
+        # 创建对话会话
+        characters = [char1_name, char2_name]
+        worldview = kb.worldview_data.get('worldview_name', '未知世界观') if kb.worldview_data else '未知世界观'
+        
+        session_id = conv_mgr.create_session(characters, worldview)
+        print(f"✓ 创建对话会话: {session_id}")
+        
+        # 交互式对话循环
+        print(f"\n=== 双AI模型对话系统 ===")
+        print(f"角色: {char1_name} vs {char2_name}")
+        print(f"世界观: {worldview}")
+        print("输入 'quit' 退出对话")
+        print("-" * 50)
+        
+        while True:
+            try:
+                # 获取用户输入
+                user_input = input("\n请输入对话主题或指令: ").strip()
+                
+                if user_input.lower() == 'quit':
+                    print("退出双AI对话系统")
+                    break
+                
+                if not user_input:
+                    print("请输入有效的对话主题")
+                    continue
+                
+                # 询问对话轮数
+                turns_input = input("请输入对话轮数 (默认4): ").strip()
+                turns = int(turns_input) if turns_input.isdigit() else 4
+                
+                # 询问历史上下文设置
+                history_input = input("使用历史对话轮数 (默认2): ").strip()
+                history_count = int(history_input) if history_input.isdigit() else 2
+                
+                context_input = input("使用上下文信息数量 (默认10): ").strip()
+                context_info_count = int(context_input) if context_input.isdigit() else 10
+                
+                print(f"\n开始对话 - 主题: {user_input}")
+                print(f"轮数: {turns}, 历史: {history_count}, 上下文: {context_info_count}")
+                print("-" * 50)
+                
+                # 运行双模型对话
+                dialogue_engine.run_dual_model_conversation(
+                    session_id, user_input, turns, history_count, context_info_count
+                )
+                
+                print("-" * 50)
+                print("对话完成！")
+                
+            except KeyboardInterrupt:
+                print("\n\n用户中断对话")
+                break
+            except Exception as e:
+                print(f"对话过程中出现错误: {e}")
+                import traceback
+                traceback.print_exc()
+        
+    
+            
+    except Exception as e:
+        print(f"✗ 对话系统启动失败: {e}")
+        import traceback
+        traceback.print_exc()
+
+def create_demo_scenario():
+    """创建演示场景"""
+    print("\n创建演示对话场景...")
+    
+    try:
+        from dual_ai_dialogue_system import RAGKnowledgeBase, ConversationManager, DualAIDialogueEngine
+        from npc_dialogue_generator import NPCDialogueGenerator
+        
+        # 初始化组件
+        kb = RAGKnowledgeBase("./knowledge_base")
+        conv_mgr = ConversationManager("./conversation_data/demo_conversations.db")
+        
+        # 检查模型路径
+        base_model_path = '/mnt/e/AI/Project02/AITrain/Qwen/Qwen3-4B'
+        lora_model_path = './output/NPC_Dialogue_LoRA/final_model'
+        
+        if not os.path.exists(base_model_path):
+            print(f"✗ 基础模型路径不存在: {base_model_path}")
+            print("请修改 main_controller.py 中的模型路径")
+            return
+        
+        if not os.path.exists(lora_model_path):
+            lora_model_path = None
+            print("⚠ LoRA模型不存在，使用基础模型")
+        
+        llm_generator = NPCDialogueGenerator(base_model_path, lora_model_path, kb.character_data)
+        dialogue_engine = DualAIDialogueEngine(kb, conv_mgr, llm_generator)
+        
+        # 创建演示对话
+        characters = ["维多利亚·布莱克伍德", "阿奇博尔德·韦恩"]
+        worldview = "克苏鲁的呼唤"
+        
+        session_id = conv_mgr.create_session(characters, worldview)
+        print(f"✓ 创建演示会话: {session_id}")
+        
+        # 运行几轮对话
+        topic = "最近发生的神秘事件"
+        print(f"\n开始演示对话 - 主题: {topic}")
+        print("-" * 40)
+        
+        # 演示不同的历史上下文设置
+        # print("演示1: 使用默认上下文设置（历史3轮，信息2个）")
+        # dialogue_engine.run_conversation_turn(session_id, characters, 6, topic)
+        
+     
+        session_id = conv_mgr.create_session(characters, worldview)
+        print(f"✓ 创建演示会话: {session_id}")
+        print("\n演示3: 使用最少历史上下文（历史1轮，信息1个）")
+        dialogue_engine.run_conversation_turn(session_id, characters, 6, topic, 1, 10)
+
+        session_id = conv_mgr.create_session(characters, worldview)
+        print(f"✓ 创建演示会话: {session_id}")
+        print("\n演示2: 使用更多历史上下文（历史10轮，信息10个）")
+        dialogue_engine.run_conversation_turn(session_id, characters, 6, topic, 5, 10)
+        
+        print(f"\n✓ 演示完成！会话ID: {session_id}")
+        print("你可以通过主对话系统继续这个对话")
+        
+    except Exception as e:
+        print(f"✗ 演示场景创建失败: {e}")
+        import traceback
+        traceback.print_exc()
+
+def show_system_status():
+    """显示系统状态"""
+    print("\n" + "="*60)
+    print("系统状态检查")
+    print("="*60)
+    
+    # 检查文件
+    files_to_check = [
+        ("./knowledge_base/worldview_template_coc.json", "世界观模板"),
+        ("./knowledge_base/character_template_detective.json", "侦探角色"),
+        ("./knowledge_base/character_template_professor.json", "教授角色"),
+        ("./pdf_to_rag_processor.py", "PDF处理器"),
+        ("./dual_ai_dialogue_system.py", "对话系统"),
+        ("./npc_dialogue_generator.py", "NPC生成器")
+    ]
+    
+    print("\n文件检查:")
+    for file_path, description in files_to_check:
+        if os.path.exists(file_path):
+            print(f"✓ {description}: {file_path}")
+        else:
+            print(f"✗ {description}: {file_path} (不存在)")
+    
+    # 检查目录
+    print("\n目录检查:")
+    directories = ["./knowledge_base", "./rag_knowledge", "./conversation_data"]
+    for dir_path in directories:
+        if os.path.exists(dir_path):
+            file_count = len([f for f in os.listdir(dir_path) if os.path.isfile(os.path.join(dir_path, f))])
+            print(f"✓ {dir_path}: {file_count} 个文件")
+        else:
+            print(f"✗ {dir_path}: 不存在")
+    
+    # 检查对话会话
+    try:
+        from dual_ai_dialogue_system import ConversationManager
+        conv_mgr = ConversationManager("./conversation_data/conversations.db")
+        sessions = conv_mgr.list_sessions()
+        print(f"\n✓ 对话会话: {len(sessions)} 个")
+    except Exception as e:
+        print(f"\n✗ 对话会话检查失败: {e}")
+
+def main():
+    """主控制程序"""
+    print("="*70)
+    print("         双AI角色对话系统 - 主控制程序")
+    print("         基于RAG的世界观增强对话引擎")
+    print("="*70)
+    
+    # 检查依赖
+    if not check_dependencies():
+        return
+    
+    # 设置目录
+    # setup_directories()
+    # copy_demo_files()
+    
+    while True:
+        print("\n" + "="*50)
+        print("主菜单 - 请选择操作:")
+        print("1. 处理PDF世界观文档 (转换为RAG格式)")
+        print("2. 查看角色设定信息")
+        print("3. 启动双AI对话系统 (支持双模型对话)")
+        print("4. 创建演示对话场景")
+        print("5. 系统状态检查")
+        print("6. 查看使用说明")
+        print("0. 退出")
+        print("="*50)
+        
+        choice = input("请输入选择 (0-6): ").strip()
+        
+        if choice == '0':
+            print("\n感谢使用双AI角色对话系统！")
+            break
+            
+        elif choice == '1':
+            process_pdf_workflow()
+            
+        elif choice == '2':
+            show_character_info()
+            
+        elif choice == '3':
+            run_dialogue_system()
+            
+        elif choice == '4':
+            create_demo_scenario()
+            
+        elif choice == '5':
+            show_system_status()
+            
+        elif choice == '6':
+            show_usage_guide()
+            
+        else:
+            print("❌ 无效选择，请重新输入")
+
+def show_usage_guide():
+    """显示使用说明"""
+    print("\n" + "="*60)
+    print("系统使用说明")
+    print("="*60)
+    
+    guide = """
+🚀 快速开始:
+1. 首次使用建议先运行"创建演示对话场景"
+2. 如有PDF世界观文档，选择"处理PDF世界观文档"
+3. 通过"启动双AI对话系统"开始角色对话
+
+📁 文档格式说明:
+- 世界观文档: worldview_template_coc.json (参考COC设定)
+- 角色设定: character_template_*.json (包含详细人设)
+
+🔧 系统功能:
+- PDF自动转换为RAG知识库
+- 基于向量相似度的上下文检索
+- 持久化对话历史存储
+- 角色设定一致性保持
+
+📝 自定义角色:
+1. 参考 character_template_*.json 格式
+2. 保存到 knowledge_base/ 目录
+3. 重启对话系统加载新角色
+
+💾 对话数据:
+- 历史对话保存在 conversation_data/ 目录
+- 支持会话恢复和历史查看
+- 自动记录使用的上下文信息
+
+⚠️  注意事项:
+- 确保模型路径正确设置
+- 首次运行需要下载向量化模型
+- PDF处理需要足够内存
+"""
+    print(guide)
+
+if __name__ == '__main__':
+    main()

AITrain/npc_dialogue_generator.py

@@ -3,6 +3,7 @@
 '''
 游戏NPC角色对话生成器
 基于微调后的LoRA模型生成角色对话
+支持双模型对话系统，每个模型扮演一个角色
 '''
 
 import torch
@@ -10,8 +11,9 @@ import json
 import random
 from peft import PeftModel
 from transformers import AutoModelForCausalLM, AutoTokenizer
-from typing import Dict, List, Optional
+from typing import Dict, List, Optional, Tuple
 import platform
+import os
 
 # Windows multiprocessing兼容性修复
 if platform.system() == "Windows":
@@ -19,86 +21,96 @@ if platform.system() == "Windows":
     multiprocessing.set_start_method('spawn', force=True)
 
 class NPCDialogueGenerator:
-    def __init__(self, base_model_path: str, lora_model_path: Optional[str] = None):
+    def __init__(self, base_model_path: str, lora_model_path: Optional[str] = None, external_character_data: Optional[Dict] = None):
         """
         初始化NPC对话生成器
         
         Args:
             base_model_path: 基础模型路径
             lora_model_path: LoRA模型路径（可选）
+            external_character_data: 外部角色数据（可选，优先使用）
         """
         self.base_model_path = base_model_path
         self.lora_model_path = lora_model_path
         self.model = None
         self.tokenizer = None
-        self.character_profiles = self._load_character_profiles()
+        
+        # 优先使用外部角色数据，如果没有则使用内置数据
+        if external_character_data:
+            self.character_profiles = self._process_external_character_data(external_character_data)
+            print(f"✓ 使用外部角色数据: {list(self.character_profiles.keys())}")
         
         self._load_model()
     
-    def _load_character_profiles(self) -> Dict:
-        """加载角色画像数据"""
-        return {
-            "维多利亚·布莱克伍德": {
-                "name": "维多利亚·布莱克伍德",
-                "title": "神秘学专家",
-                "personality": ["理性分析", "谨慎小心", "实用主义", "思维缜密"],
-                "background": "拥有丰富神秘学知识和战斗经验的侦探，既是非凡者也是夏洛克·莫里亚蒂",
-                "speech_patterns": ["会使用专业术语", "经常进行逻辑分析", "对危险保持警告", "内心独白较多"],
-                "sample_dialogues": [
-                    "好奇往往是导致死亡的主要因素。",
-                    "总之，我的任务到此为止。",
-                    "这需要仔细分析才能得出结论。"
-                ]
-            },
-            "阿奇博尔德·韦恩博士": {
-                "name": "阿奇博尔德·韦恩博士",
-                "title": "神秘学导师",
-                "personality": ["沉稳睿智", "言简意赅", "关怀学生", "经验丰富"],
-                "background": "神秘学领域的资深专家，经验极其丰富的导师，知识渊博",
-                "speech_patterns": ["话语简练但信息量大", "给予实用指导", "语调平和但权威", "关心但保持距离"],
-                "sample_dialogues": [
-                    "耐心是修炼的基础。",
-                    "不要急于求成，稳扎稳打比什么都重要。",
-                    "这种情况需要格外小心。"
-                ]
-            },
-            "塔利姆": {
-                "name": "塔利姆",
-                "title": "文雅绅士",
-                "personality": ["礼貌尊敬", "有文化素养", "寻求帮助", "温和友善"],
-                "background": "受过良好教育的普通人，有一定的文学修养，遇到困难时会寻求专家帮助",
-                "speech_patterns": ["使用礼貌称谓", "表达困惑时措辞文雅", "会引用文学作品", "语气温和"],
-                "sample_dialogues": [
-                    "噢，尊敬的大侦探，你最近在忙碌什么？",
-                    "这不是《罗密欧与朱丽叶》的故事！",
-                    "我有个朋友遇到了困难..."
-                ]
-            },
-            "艾伦": {
-                "name": "艾伦",
-                "title": "困扰的求助者",
-                "personality": ["焦虑不安", "详细描述", "半信半疑", "急需帮助"],
-                "background": "普通人，但最近遭遇了一系列神秘的厄运事件，怀疑受到诅咒",
-                "speech_patterns": ["情绪紧张", "会详细描述遭遇", "语气急切", "表现出恐惧"],
-                "sample_dialogues": [
-                    "最近我总是遭遇各种厄运...",
-                    "我怀疑是不是受到了什么诅咒。",
-                    "请帮帮我，我不知道该怎么办！"
-                ]
-            },
-            "戴莉.西蒙妮": {
-                "name": "戴莉·西蒙妮",
-                "title": "专业调查员",
-                "personality": ["专业简洁", "直接明确", "严谨认真", "目标导向"],
-                "background": "负责调查神秘事件的专业人员，办事效率高，问题直接",
-                "speech_patterns": ["问题直接明确", "语气专业", "注重事实", "简洁有力"],
-                "sample_dialogues": [
-                    "请详细描述事件经过。",
-                    "有什么证据可以证明？",
-                    "这件事需要立即调查。"
-                ]
+    def _process_external_character_data(self, external_data: Dict) -> Dict:
+        """
+        处理外部角色数据，转换为对话生成器可用的格式
+        
+        Args:
+            external_data: 来自knowledge_base的角色数据
+            
+        Returns:
+            处理后的角色数据字典
+        """
+        processed_profiles = {}
+        
+        for char_name, char_data in external_data.items():
+            # 提取基本信息
+            basic_info = char_data.get('basic_info', {})
+            personality = char_data.get('personality', {})
+            background = char_data.get('background', {})
+            skills = char_data.get('skills_and_abilities', {})
+            speech_patterns = char_data.get('speech_patterns', {})
+            
+            # 构建角色画像
+            profile = {
+                "name": char_data.get('character_name', char_name),
+                "title": basic_info.get('occupation', '未知'),
+                "personality": personality.get('core_traits', []) + personality.get('strengths', []),
+                "background": background.get('childhood', '') + ' ' + background.get('education', ''),
+                "speech_patterns": speech_patterns.get('vocabulary', []) + speech_patterns.get('tone', []),
+                "sample_dialogues": self._generate_sample_dialogues(char_data),
+                # 保存完整数据供高级功能使用
+                "full_data": char_data
             }
-        }
+            
+            processed_profiles[char_name] = profile
+        
+        return processed_profiles
+    
+    def _generate_sample_dialogues(self, char_data: Dict) -> List[str]:
+        """
+        基于角色数据生成示例对话
+        
+        Args:
+            char_data: 角色数据
+            
+        Returns:
+            示例对话列表
+        """
+        # 这里可以根据角色的性格、背景等生成更合适的示例对话
+        # 暂时返回一些通用的示例
+        basic_info = char_data.get('basic_info', {})
+        occupation = basic_info.get('occupation', '角色')
+        
+        if '侦探' in occupation or '调查员' in occupation:
+            return [
+                "我需要仔细分析这个案件。",
+                "每个细节都可能很重要。",
+                "让我重新梳理一下线索。"
+            ]
+        elif '教授' in occupation or '博士' in occupation:
+            return [
+                "根据我的研究，这个现象很特殊。",
+                "我们需要更谨慎地处理这个问题。",
+                "知识就是力量，但也要小心使用。"
+            ]
+        else:
+            return [
+                "我遇到了一些困难。",
+                "请帮帮我。",
+                "这太奇怪了。"
+            ]
     
     def _load_model(self):
         """加载模型和分词器"""
@@ -121,18 +133,19 @@ class NPCDialogueGenerator:
         )
         
         # 如果有LoRA模型，则加载
-        if self.lora_model_path:
-            print(f"Loading LoRA weights from: {self.lora_model_path}")
-            self.model = PeftModel.from_pretrained(self.model, self.lora_model_path)
+        # if self.lora_model_path:
+        #     print(f"Loading LoRA weights from: {self.lora_model_path}")
+        #     self.model = PeftModel.from_pretrained(self.model, self.lora_model_path)
        
     def generate_character_dialogue(
         self, 
         character_name: str, 
         context: str = "", 
-        user_input: str = "", 
         temperature: float = 0.8,
         max_new_tokens: int = 150,
-        top_p: float = 0.9
+        top_p: float = 0.9,
+        dialogue_history: List[Dict] = None,
+        history_context_count: int = 3
     ) -> str:
         """
         生成指定角色的对话
@@ -144,6 +157,8 @@ class NPCDialogueGenerator:
             temperature: 采样温度
             max_new_tokens: 最大生成token数
             top_p: 核采样参数
+            dialogue_history: 对话历史记录列表，每个元素包含speaker和content
+            history_context_count: 使用的历史对话轮数（默认3轮）
             
         Returns:
             生成的对话内容
@@ -153,12 +168,12 @@ class NPCDialogueGenerator:
         
         profile = self.character_profiles[character_name]
         
-        # 构建系统提示
-        system_prompt = self._build_system_prompt(profile, context)
+        # 构建系统提示，包含历史对话数据
+        system_prompt = self._build_system_prompt(profile, context, dialogue_history, history_context_count)
         
         # 构建用户输入
-        if not user_input:
-            user_input = "请说一段符合你角色设定的话。"
+        user_input = "请说一段符合你角色设定的话,保持对话的连贯性。"
+            
         
         # 准备消息
         messages = [
@@ -178,6 +193,16 @@ class NPCDialogueGenerator:
         
         # 移动到设备
         inputs = {k: v.to(self.model.device) for k, v in inputs.items()}
+        # 计算input token数并与模型最大token数比较
+        input_token_count = inputs['input_ids'].shape[1]
+        try:
+            max_model_tokens = self.model.config.max_position_embeddings
+        except AttributeError:
+            max_model_tokens = 2048
+        
+        if input_token_count + max_new_tokens > max_model_tokens:
+            print(f"警告：当前输入token数({input_token_count})加上最大生成token数({max_new_tokens})超过模型最大token数({max_model_tokens})，可能导致生成结果不完整或报错。")
+        
       
         # 生成对话
         with torch.no_grad():
@@ -185,10 +210,10 @@ class NPCDialogueGenerator:
                 **inputs,
                 max_new_tokens=max_new_tokens,
                 do_sample=True,
-                temperature=temperature,
-                top_p=top_p,
+                temperature=0.95,
+                top_p=0.92,
                 pad_token_id=self.tokenizer.eos_token_id,
-                repetition_penalty=1.1
+                repetition_penalty=1.15
             )
         
         # 解码输出
@@ -197,8 +222,15 @@ class NPCDialogueGenerator:
         
         return dialogue
     
-    def _build_system_prompt(self, profile: Dict, context: str = "") -> str:
-        """构建系统提示"""
+    def _build_system_prompt(self, profile: Dict, context: str = "", dialogue_history: List[Dict] = None, history_context_count: int = 3) -> str:
+        """构建系统提示
+        
+        Args:
+            profile: 角色配置信息
+            context: 当前情境
+            dialogue_history: 对话历史记录列表，每个元素包含speaker和content
+            history_context_count: 使用的历史对话轮数（默认3轮）
+        """
         personality_str = "、".join(profile["personality"])
         speech_pattern_str = "；".join(profile["speech_patterns"])
         
@@ -207,9 +239,24 @@ class NPCDialogueGenerator:
                     性格特点：{personality_str}
                     说话风格：{speech_pattern_str}
                     请严格按照这个角色的设定来回应，保持角色的一致性和独特性。"""
+        
+        # 添加当前情境
         if context:
             system_prompt += f"\n\n当前情境：{context}"
-            return system_prompt
+        
+        # 添加历史对话数据（参考generate_character_prompt的实现）
+        if dialogue_history:
+            system_prompt += "\n\n最近的对话："
+            # 使用参数控制历史对话轮数
+            history_to_use = dialogue_history[-history_context_count:] if history_context_count > 0 else []
+            for turn in history_to_use:
+                system_prompt += f"{turn.speaker}: {turn.content}"
+                # speaker = turn.get('speaker', '未知')
+                # content = turn.get('content', '')
+                # if content:
+                #     system_prompt += f"\n{speaker}: {content}"
+        
+        return system_prompt
 
     def generate_dialogue_conversation(self, character1: str, character2: str, topic: str, turns: int = 4) -> List[Dict]:
         """生成两个角色之间的对话
@@ -261,102 +308,264 @@ class NPCDialogueGenerator:
         """列出所有可用角色"""
         return list(self.character_profiles.keys())
 
-    def main():
-        """测试对话生成器"""
-        # 配置路径
-        base_model_path = '/mnt/g/Project02/AITrain/Qwen/Qwen3-8B-AWQ'
-        lora_model_path = './output/NPC_Dialogue_LoRA/final_model'  # 如果没有训练LoRA，设为None
+class DualModelDialogueGenerator:
+    """双模型对话生成器 - 每个模型扮演一个角色"""
     
-        # 检查LoRA模型是否存在
-        import os
-        if not os.path.exists(lora_model_path):
-            print("LoRA模型不存在，使用基础模型")
-            lora_model_path = None
+    def __init__(self, 
+                 base_model_path: str,
+                 character1_config: Dict,
+                 character2_config: Dict,
+                 lora_model_path: Optional[str] = None):
+        """
+        初始化双模型对话生成器
         
-        # 创建对话生成器
-        generator = NPCDialogueGenerator(base_model_path, lora_model_path)
+        Args:
+            base_model_path: 基础模型路径
+            character1_config: 角色1配置 {"name": "角色名", "lora_path": "LoRA路径", "character_data": 角色数据}
+            character2_config: 角色2配置 {"name": "角色名", "lora_path": "LoRA路径", "character_data": 角色数据}
+            lora_model_path: 通用LoRA模型路径（可选）
+        """
+        self.base_model_path = base_model_path
+        self.character1_config = character1_config
+        self.character2_config = character2_config
         
-        print("=== 游戏NPC角色对话生成器 ===")
-        print(f"可用角色：{', '.join(generator.list_available_characters())}")
+        # 为每个角色创建独立的模型实例
+        self.character1_generator = None
+        self.character2_generator = None
         
-        # 测试单个角色对话生成
-        print("\n=== 单角色对话测试 ===")
-        test_scenarios = [
-            {
-                "character": "克莱恩",
-                "context": "玩家向你咨询神秘学知识",
-                "input": "请告诉我一些关于灵界的注意事项。"
-            },
-            {
-                "character": "阿兹克",
-                "context": "学生遇到了修炼瓶颈",
-                "input": "导师，我在修炼中遇到了困难。"
-            },
-            {
-                "character": "塔利姆",
-                "context": "在俱乐部偶遇老朋友",
-                "input": "好久不见，最近怎么样？"
-            }
-        ]
+        self._initialize_character_models()
     
-        for scenario in test_scenarios:
-            print(f"\n--- {scenario['character']} ---")
-            print(f"情境：{scenario['context']}")
-            print(f"输入：{scenario['input']}")
+    def _initialize_character_models(self):
+        """初始化两个角色的模型"""
+        print("=== 初始化双模型对话系统 ===")
         
-            dialogue = generator.generate_character_dialogue(
-                scenario["character"],
-                scenario["context"],
-                scenario["input"]
-            )
-            print(f"回复：{dialogue}")
-        
-        # 测试角色间对话
-        print("\n=== 角色间对话测试 ===")
-        conversation = generator.generate_dialogue_conversation(
-            "克莱恩", "塔利姆", "最近遇到的神秘事件", turns=4
+        # 初始化角色1的模型
+        print(f"\n初始化角色1: {self.character1_config['name']}")
+        char1_lora_path = self.character1_config.get('lora_path') or self.character1_config.get('lora_model_path')
+        self.character1_generator = NPCDialogueGenerator(
+            self.base_model_path,
+            char1_lora_path,
+            {self.character1_config['name']: self.character1_config['character_data']}
         )
         
-        for turn in conversation:
-            print(f"{turn['speaker']}：{turn['dialogue']}")
+        # 初始化角色2的模型
+        print(f"\n初始化角色2: {self.character2_config['name']}")
+        char2_lora_path = self.character2_config.get('lora_path') or self.character2_config.get('lora_model_path')
+        self.character2_generator = NPCDialogueGenerator(
+            self.base_model_path,
+            char2_lora_path,
+            {self.character2_config['name']: self.character2_config['character_data']}
+        )
         
-        # 交互式对话模式
-        print("\n=== 交互式对话模式 ===")
-        print("输入格式：角色名 上下文 用户输入")
-        print("例如：克莱恩 在俱乐部 请给我一些建议")
-        print("输入'quit'退出")
+        print("✓ 双模型对话系统初始化完成")
     
-        while True:
-            try:
-                user_command = input("\n请输入指令: ").strip()
-                if user_command.lower() == 'quit':
-                    break
+    def generate_dual_character_dialogue(self, 
+                                       character_name: str,
+                                       context: str = "",
+                                       temperature: float = 0.8,
+                                       max_new_tokens: int = 150,
+                                       dialogue_history: str = "",
+                                       history_context_count: int = 3) -> str:
+        """
+        生成指定角色的对话（使用对应的模型）
         
-                parts = user_command.split(' ', 2)
-                if len(parts) < 2:
-                    print("格式错误，请使用：角色名 上下文 [用户输入]")
-                    continue
+        Args:
+            character_name: 角色名称
+            context: 对话上下文
+            user_input: 用户输入
+            temperature: 采样温度
+            max_new_tokens: 最大生成token数
             
-                character = parts[0]
-                context = parts[1]
-                user_input = parts[2] if len(parts) > 2 else ""
+        Returns:
+            生成的对话内容
+        """
+        if character_name == self.character1_config['name']:
+            return self.character1_generator.generate_character_dialogue(
+                character_name, context,  temperature, max_new_tokens, dialogue_history = dialogue_history, history_context_count=history_context_count,
+            )
+        elif character_name == self.character2_config['name']:
+            return self.character2_generator.generate_character_dialogue(
+                character_name, context,  temperature, max_new_tokens, dialogue_history = dialogue_history, history_context_count=history_context_count,
+            )
+        else:
+            raise ValueError(f"Unknown character: {character_name}")
     
-                if character not in generator.list_available_characters():
-                    print(f"未知角色：{character}")
-                    print(f"可用角色：{', '.join(generator.list_available_characters())}")
-                    continue
+    def run_dual_character_conversation(self, 
+                                      topic: str = "",
+                                      turn_index: int = 4,
+                                      context: str = "",
+                                      dialogue_history: str = "",
+                                      history_context_count: int = 3,
+                                      temperature: float = 0.8,
+                                      max_new_tokens: int = 150) -> List[Dict]:
+        """
+        运行双角色对话
         
-                dialogue = generator.generate_character_dialogue(
-                    character, context, user_input
-                )
-                print(f"\n{character}：{dialogue}")
+        Args:
+            topic: 对话主题
+            turns: 对话轮数
+            context: 额外上下文
+            temperature: 采样温度
+            max_new_tokens: 最大生成token数
             
-            except KeyboardInterrupt:
-                break
-            except Exception as e:
-                print(f"生成对话时出错：{e}")
+        Returns:
+            对话列表
+        """
+        conversation = []
+        char1_name = self.character1_config['name']
+        char2_name = self.character2_config['name']
         
-        print("\n对话生成器已退出")
+        # 构建完整上下文
+        full_context = f"现在{char1_name}和{char2_name}在讨论关于{topic}的话题。{context}"
         
-    if __name__ == '__main__':
-        main()
+        # print(f"\n=== 开始双角色对话 ===")
+        # print(f"主题: {topic}")
+        # print(f"角色: {char1_name} vs {char2_name}")
+        # print("-" * 50)
+        
+
+        if turn_index % 2 == 0:
+            # 角色1说话
+            speaker = char1_name
+            # if turn_index == 0:
+            #     user_input = f"开始和{char2_name}讨论{topic}这个话题。"
+            # else:
+            #     last_dialogue = conversation[-1]["dialogue"]
+            #     user_input = f"{char2_name}刚才说：\"{last_dialogue}\"。请回应。"
+        else:
+            # 角色2说话
+            speaker = char2_name
+            # last_dialogue = conversation[-1]["dialogue"]
+            # user_input = f"{char1_name}刚才说：\"{last_dialogue}\"。请回应。"
+        
+        print(f"\n[第{turn_index+1}轮] {speaker}正在思考...")
+        
+        # 使用对应角色的模型生成对话
+        dialogue = self.generate_dual_character_dialogue(
+            speaker, full_context,  temperature, max_new_tokens, dialogue_history, history_context_count
+        )
+        
+        conversation.append({
+            "turn": turn_index + 1,
+            "speaker": speaker,
+            "dialogue": dialogue,
+            "context_used": full_context[:100] + "..." if len(full_context) > 100 else full_context
+        })
+        
+        print(f"{speaker}: {dialogue}")
+        
+      
+        
+        return conversation
+    
+    def get_character_info(self, character_name: str) -> Dict:
+        """获取角色信息"""
+        if character_name == self.character1_config['name']:
+            return self.character1_generator.get_character_info(character_name)
+        elif character_name == self.character2_config['name']:
+            return self.character2_generator.get_character_info(character_name)
+        else:
+            return {}
+    
+    def list_characters(self) -> List[str]:
+        """列出两个角色名称"""
+        return [self.character1_config['name'], self.character2_config['name']]
+
+# def main():
+#     """测试对话生成器"""
+#     # 配置路径
+#     base_model_path = '/mnt/g/Project02/AITrain/Qwen/Qwen3-8B-AWQ'
+#     lora_model_path = './output/NPC_Dialogue_LoRA/final_model'  # 如果没有训练LoRA，设为None
+    
+#     # 检查LoRA模型是否存在
+#     if not os.path.exists(lora_model_path):
+#         print("LoRA模型不存在，使用基础模型")
+#         lora_model_path = None
+    
+#     # 创建对话生成器
+#     generator = NPCDialogueGenerator(base_model_path, lora_model_path)
+    
+#     print("=== 游戏NPC角色对话生成器 ===")
+#     print(f"可用角色：{', '.join(generator.list_available_characters())}")
+    
+#     # 测试单个角色对话生成
+#     print("\n=== 单角色对话测试 ===")
+#     test_scenarios = [
+#         {
+#             "character": "克莱恩",
+#             "context": "玩家向你咨询神秘学知识",
+#             "input": "请告诉我一些关于灵界的注意事项。"
+#         },
+#         {
+#             "character": "阿兹克",
+#             "context": "学生遇到了修炼瓶颈",
+#             "input": "导师，我在修炼中遇到了困难。"
+#         },
+#         {
+#             "character": "塔利姆",
+#             "context": "在俱乐部偶遇老朋友",
+#             "input": "好久不见，最近怎么样？"
+#         }
+#     ]
+    
+#     for scenario in test_scenarios:
+#         print(f"\n--- {scenario['character']} ---")
+#         print(f"情境：{scenario['context']}")
+#         print(f"输入：{scenario['input']}")
+        
+#         dialogue = generator.generate_character_dialogue(
+#             scenario["character"],
+#             scenario["context"],
+#             scenario["input"]
+#         )
+#         print(f"回复：{dialogue}")
+    
+#     # 测试角色间对话
+#     print("\n=== 角色间对话测试 ===")
+#     conversation = generator.generate_dialogue_conversation(
+#         "克莱恩", "塔利姆", "最近遇到的神秘事件", turns=4
+#     )
+    
+#     for turn in conversation:
+#         print(f"{turn['speaker']}：{turn['dialogue']}")
+    
+#     # 交互式对话模式
+#     print("\n=== 交互式对话模式 ===")
+#     print("输入格式：角色名 上下文 用户输入")
+#     print("例如：克莱恩 在俱乐部 请给我一些建议")
+#     print("输入'quit'退出")
+    
+#     while True:
+#         try:
+#             user_command = input("\n请输入指令: ").strip()
+#             if user_command.lower() == 'quit':
+#                 break
+            
+#             parts = user_command.split(' ', 2)
+#             if len(parts) < 2:
+#                 print("格式错误，请使用：角色名 上下文 [用户输入]")
+#                 continue
+            
+#             character = parts[0]
+#             context = parts[1]
+#             user_input = parts[2] if len(parts) > 2 else ""
+            
+#             if character not in generator.list_available_characters():
+#                 print(f"未知角色：{character}")
+#                 print(f"可用角色：{', '.join(generator.list_available_characters())}")
+#                 continue
+            
+#             dialogue = generator.generate_character_dialogue(
+#                 character, context, user_input
+#             )
+#             print(f"\n{character}：{dialogue}")
+            
+#         except KeyboardInterrupt:
+#             break
+#         except Exception as e:
+#             print(f"生成对话时出错：{e}")
+    
+#     print("\n对话生成器已退出")
+
+# if __name__ == '__main__':
+#     main()

AITrain/requirements.txt

@@ -16,5 +16,6 @@ datasets==3.5.1
 peft==0.15.2
 autoawq
 PyPDF2
-sentence-transformersfaiss-cpu
+sentence-transformers
+faiss-cpu
 fitz