v0.0.4-debug

- 优化hybrid_search_recall tokp = 8

config/config.ini

@@ -81,7 +81,7 @@ QWEN_LOCAL_1_5B_MODEL_ID=Qwen3-4B
 QWEN_LOCAL_1_5B_API_KEY=dummy
 
 [rerank_model]
-BGE_RERANKER_SERVER_RUL=http://192.168.91.253:9005/v1/rerank
+BGE_RERANKER_SERVER_RUL=http://192.168.91.254:9005/v1/rerank
 BGE_RERANKER_MODEL_ID=BAAI/bge-reranker-v2-m3
 BGE_RERANKER_API_KEY=dummy
 BGE_RERANKER_TOP_N=10

core/construction_review/component/ai_review_engine.py

@@ -74,6 +74,7 @@ class Stage(Enum):
     BASIC = {
         'reviewer_type':'basic',
         'grammar': 'grammar_check',
+        'sensitive':'sensitive_check',
         'semantic': 'semantic_logic_check',
         'completeness': 'completeness_check',
         'timeliness': 'timeliness_check',

foundation/ai/models/rerank_model.py

@@ -42,7 +42,7 @@ class LqReranker:
                 top_k = self.top_k
             
 
-            server_logger.info(f"开始执行重排序，查询:, 候选文档数量: {len(candidates)}")
+            server_logger.info(f"开始执行重排序，查询: '{query}', 候选文档数量: {len(candidates)}")
 
             # 构建重排序请求
             rerank_request = {

foundation/ai/rag/retrieval/retrieval.py

@@ -22,7 +22,7 @@ class RetrievalManager:
         self.sparse_weight = config_handler.get('hybrid_search', 'SPARSE_WEIGHT', 0.3)
 
     def hybrid_search_recall(self, collection_name: str, query_text: str,
-                           top_k: int = 10, ranker_type: str = "weighted",
+                           top_k: int = 8, ranker_type: str = "weighted",
                            dense_weight: float = 0.7, sparse_weight: float = 0.3) -> List[Dict[str, Any]]:
         """
         混合搜索召回 - 向量+BM25召回

test/batch_multi_stage_recall.py

@@ -1,171 +0,0 @@
-#!/usr/bin/env python
-# -*- coding: utf-8 -*-
-"""
-批量查询多阶段召回示例脚本
-
-用途：
-- 复用 Milvus 混合检索 + rerank 的多阶段召回能力
-- 一次传入多条查询，逐条输出召回摘要，便于对比效果
-
-运行：
-    python test/batch_multi_stage_recall.py
-"""
-
-import os
-import sys
-import time
-from typing import List, Dict, Any
-
-# 确保可以从项目根目录导入
-sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-
-from foundation.ai.rag.retrieval.retrieval import retrieval_manager
-from foundation.database.base.vector.milvus_vector import MilvusVectorManager
-
-
-def build_sample_documents() -> List[Dict[str, Any]]:
-    """构造示例文档，便于快速跑通脚本。"""
-    return [
-        {
-            "content": "大模型是一类具有大量参数的人工智能模型，通常拥有数十亿到数千亿个参数。",
-            "metadata": {"source": "ai_basics", "category": "technology"},
-        },
-        {
-            "content": "深度学习是机器学习的一个分支，使用多层神经网络来学习数据的复杂模式。",
-            "metadata": {"source": "ml_basics", "category": "technology"},
-        },
-        {
-            "content": "自然语言处理是人工智能的一个重要领域，专注于计算机与人类语言的交互。",
-            "metadata": {"source": "nlp_basics", "category": "ai_field"},
-        },
-        {
-            "content": "Transformer架构是现代大语言模型的基础，由Google在2017年提出。",
-            "metadata": {"source": "transformer_info", "category": "architecture"},
-        },
-        {
-            "content": "GPT、BERT、T5等都是基于Transformer架构的著名预训练语言模型。",
-            "metadata": {"source": "model_examples", "category": "models"},
-        },
-        {
-            "content": "苹果公司是一家美国跨国技术公司，总部位于加利福尼亚州库比蒂诺。",
-            "metadata": {"source": "company_info", "category": "business"},
-        },
-        {
-            "content": "机器学习算法需要大量的训练数据来学习有效的特征表示。",
-            "metadata": {"source": "ml_training", "category": "technology"},
-        },
-        {
-            "content": "计算机视觉是人工智能的另一个重要分支，专注于图像和视频的理解。",
-            "metadata": {"source": "cv_basics", "category": "ai_field"},
-        },
-        {
-            "content": "强化学习通过与环境交互来学习最优策略，广泛应用于游戏和机器人控制。",
-            "metadata": {"source": "rl_basics", "category": "technology"},
-        },
-        {
-            "content": "预训练-微调范式已经成为大模型训练的标准方法，可以显著提高模型性能。",
-            "metadata": {"source": "training_paradigm", "category": "technology"},
-        },
-    ]
-
-
-def create_collection_with_samples(collection_name: str) -> str:
-    """创建（或覆盖）测试集合并写入示例文档。"""
-    docs = build_sample_documents()
-    vector_manager = MilvusVectorManager()
-    vectorstore = vector_manager.create_hybrid_collection(
-        collection_name=collection_name,
-        documents=docs,
-    )
-    if vectorstore:
-        print(f"✓ 集合 '{collection_name}' 准备就绪，文档数: {len(docs)}")
-        return collection_name
-    raise RuntimeError("创建或写入集合失败")
-
-
-def run_batch_queries(
-    collection_name: str,
-    queries: List[str],
-    hybrid_top_k: int = 10,
-    final_top_k: int = 5,
-) -> List[Dict[str, Any]]:
-    """
-    对多条查询依次执行多阶段召回。
-    返回所有结果（按查询顺序拼接）。
-    """
-    all_results: List[Dict[str, Any]] = []
-    for idx, query in enumerate(queries, 1):
-        print("\n" + "-" * 60)
-        print(f"[{idx}/{len(queries)}] 查询: {query}")
-        start = time.time()
-        try:
-            results = retrieval_manager.multi_stage_recall(
-                collection_name=collection_name,
-                query_text=query,
-                hybrid_top_k=hybrid_top_k,
-                final_top_k=final_top_k,
-            )
-        except Exception as exc:
-            print(f"✗ 召回失败: {exc}")
-            continue
-
-        cost = time.time() - start
-        print(f"✓ 召回完成，耗时 {cost:.2f}s，返回 {len(results)} 条")
-
-        # 展示前 3 条
-        for i, r in enumerate(results[:3], 1):
-            print(
-                f"  {i}. sim={r.get('similarity', 0):.4f} "
-                f"rerank={r.get('rerank_score', 0):.4f} "
-                f"hybrid={r.get('hybrid_score', 0):.4f}"
-            )
-            print(f"     内容: {r.get('text_content', '')[:80]}...")
-            print(f"     元数据: {r.get('metadata', {})}")
-
-        all_results.extend(results)
-    return all_results
-
-
-def main():
-    collection_name = "batch_bfp_collection"
-
-    # 1) 准备集合与示例数据
-    try:
-        create_collection_with_samples(collection_name)
-    except Exception as exc:
-        print(f"✗ 集合准备失败: {exc}")
-        sys.exit(1)
-
-    # 2) 定义批量查询
-    batch_queries = [
-        "什么是大模型？",
-        "Transformer架构有什么特点",
-        "苹果公司做什么的",
-        "深度学习和机器学习的关系",
-        "人工智能伦理",
-    ]
-
-    # 3) 批量召回
-    all_results = run_batch_queries(
-        collection_name=collection_name,
-        queries=batch_queries,
-        hybrid_top_k=12,
-        final_top_k=6,
-    )
-
-    # 4) 简单汇总
-    print("\n" + "=" * 60)
-    print("批量召回完成")
-    print(f"总查询数: {len(batch_queries)}")
-    print(f"累计结果数: {len(all_results)}")
-
-
-if __name__ == "__main__":
-    main()
-
-
-
-
-
-
-

test/test_rag copy.py

@@ -1,26 +0,0 @@
-import os
-import sys
-current_script_path = os.path.abspath(__file__)
-script_dir = os.path.dirname(current_script_path)
-project_root = os.path.abspath(os.path.join(script_dir, "../../"))
-if project_root not in sys.path:
-    sys.path.insert(0, project_root)
-from foundation.ai.rag.retrieval.retrieval import retrieval_manager
-
-
-query = "实体：通风孔；背景：在箱梁腹板上设置通风孔，用于结构内部空气流通，需保证与预应力钢筋保护层满足间距要求，并水平设置"
-collection = "first_bfp_collection"  # 你的 Milvus 集合名
-
-# 二阶段：Milvus混合检索(向量+BM25) → BGE重排
-results = retrieval_manager.multi_stage_recall(
-    collection_name=collection,
-    query_text=query,
-    hybrid_top_k=50,   # 第一阶段取多少候选
-    top_k=3,          # 最终返回条数
-    ranker_type="weighted"  # 或 "rrf"
-)
-
-print(results)
-for item in results:
-    print(item["rerank_score"], item["text_content"])
-    # 元数据在 item["metadata"]（来自混合检索阶段）