fix: restore main runtime sources and cache consistency

config.example.py

@@ -9,11 +9,13 @@ import os
 # 一、API 密钥与模型
 # ==============================================================================
 
-# 通义千问 LLM 服务
+# OpenAI 兼容 LLM 服务（当前默认使用小米 MiMo）
 DASHSCOPE_API_KEY = os.getenv("DASHSCOPE_API_KEY", "")
-DASHSCOPE_BASE_URL = os.getenv("DASHSCOPE_BASE_URL", "https://dashscope.aliyuncs.com/compatible-mode/v1")
-DASHSCOPE_MODEL = os.getenv("DASHSCOPE_MODEL", "qwen-flash")           # 文本生成模型
-RAG_CHAT_MODEL = os.getenv("RAG_CHAT_MODEL", "qwen-flash")             # RAG 对话模型
+DASHSCOPE_BASE_URL = os.getenv("DASHSCOPE_BASE_URL", "https://token-plan-cn.xiaomimimo.com/v1")
+DASHSCOPE_MODEL = os.getenv("DASHSCOPE_MODEL", "mimo-v2.5")             # 文本生成模型
+RAG_CHAT_MODEL = os.getenv("RAG_CHAT_MODEL", "mimo-v2.5")               # RAG 对话模型
+INTENT_MODEL = os.getenv("INTENT_MODEL", "mimo-v2.5")                   # 意图分析模型
+VLM_MODEL = os.getenv("VLM_MODEL", "mimo-v2.5")                         # 图片理解模型
 
 # 兼容旧变量名（逐步迁移到 DASHSCOPE_* 命名）
 API_KEY = DASHSCOPE_API_KEY
@@ -27,6 +29,7 @@ MODEL = DASHSCOPE_MODEL
 APP_ENV = os.getenv("APP_ENV", "dev")   # dev / prod
 IS_DEV = APP_ENV == "dev"
 IS_PROD = APP_ENV == "prod"
+DEV_MODE = os.getenv("DEV_MODE", "true").lower() != "false"
 
 # 开发/生产环境自动切换
 ENABLE_SESSION = IS_DEV          # 会话存储（仅开发环境）
@@ -85,10 +88,11 @@ RERANK_DEVICE = os.getenv("RERANK_DEVICE", os.getenv("DEVICE", "auto"))
 # ----- 通用问答 -----
 LLM_TEMPERATURE = 0.7            # 生成温度（0=确定性，1=随机性）
 LLM_MAX_TOKENS = 3000            # 最大输出 token 数
+LLM_DISABLE_THINKING = os.getenv("LLM_DISABLE_THINKING", "true").lower() != "false"
 
 # ----- 意图分析（轻量、确定性高）-----
 INTENT_TEMPERATURE = 0.1
-INTENT_MAX_TOKENS = 300
+INTENT_MAX_TOKENS = 2048
 INTENT_HISTORY_WINDOW = 6        # 分析时取最近几条历史消息
 
 # ==============================================================================
@@ -201,6 +205,9 @@ MAX_QUERY_REWRITES = 1
 
 # MinerU 解析器
 MINERU_DEVICE_MODE = os.getenv("MINERU_DEVICE_MODE", "cpu")  # cpu / cuda
+MINERU_MODEL_VERSION = os.getenv("MINERU_MODEL_VERSION", "vlm")
+MINERU_LOCAL_BACKEND = os.getenv("MINERU_LOCAL_BACKEND", "pipeline")
+MINERU_ONLINE_TIMEOUT = int(os.getenv("MINERU_ONLINE_TIMEOUT", "300"))
 
 # MinerU 在线 API（作为本地解析失败时的备选）
 MINERU_API_TOKEN = os.getenv("MINERU_API_TOKEN", "")  # 在 https://mineru.net/apiManage/token 申请
@@ -246,6 +253,20 @@ CONFIDENCE_CAUTION_THRESHOLD = 0.30 # top-3 均分低于此值时，提示 LLM
 ENUM_QUERY_DISABLE_TOPK_SHRINK = True
 ENUM_QUERY_MMR_LAMBDA = 0.85
 
+# 章节聚类提升与 BM25/CrossEncoder 分歧救援
+SECTION_CLUSTER_BOOST_ENABLED = True
+SECTION_CLUSTER_RESCUE_ENABLED = True
+BM25_DIVERGENCE_RESCUE_ENABLED = True
+BM25_DIVERGENCE_MAX_RANK = 3
+CLUSTER_MIN_MEMBERS = 3
+CLUSTER_MIN_TYPES = 2
+CLUSTER_SEED_FLOOR = 0.35
+CLUSTER_RESCUE_FLOOR = 0.06
+CLUSTER_MAX_BOOST_PER_SECTION = 8
+CLUSTER_MAX_SECTIONS = 3
+CLUSTER_MAX_RESCUE_PER_SECTION = 6
+CLUSTER_SECTION_PREFIX_LEVELS = 2
+
 # ==============================================================================
 # 十一、可选功能配置
 # ==============================================================================
@@ -261,3 +282,20 @@ def get_llm_client():
     """获取 LLM 客户端实例"""
     from openai import OpenAI
     return OpenAI(api_key=DASHSCOPE_API_KEY, base_url=DASHSCOPE_BASE_URL)
+
+
+_intent_client = None
+
+
+def get_intent_client():
+    """获取意图分析专用 LLM 客户端（进程内复用连接池）。"""
+    global _intent_client
+    if _intent_client is None:
+        if not DASHSCOPE_API_KEY:
+            raise ValueError("DASHSCOPE_API_KEY 未配置，请在环境变量中设置")
+        from openai import OpenAI
+        _intent_client = OpenAI(
+            api_key=DASHSCOPE_API_KEY,
+            base_url=DASHSCOPE_BASE_URL,
+        )
+    return _intent_client