导入方式修改

backend/app/model_services/chat_services.py

@@ -23,7 +23,7 @@ from .base import (
     FallbackServiceChain,
     SingletonServiceManager
 )
-from app.config import (
+from ..config import (
     VLLM_BASE_URL,
     LLM_API_KEY,
     ZHIPUAI_API_KEY,
@@ -203,7 +203,7 @@ class LocalSmallModelProvider(BaseServiceProvider[BaseChatModel]):
     """
     
     def __init__(self, model: str = None):
-        from app.config import SMALL_LOCAL_MODEL_NAME, SMALL_VLLM_BASE_URL, SMALL_LLM_API_KEY
+        from ..config import SMALL_LOCAL_MODEL_NAME, SMALL_VLLM_BASE_URL, SMALL_LLM_API_KEY
         super().__init__("local_small")
         self._model = model or SMALL_LOCAL_MODEL_NAME
         self._base_url = SMALL_VLLM_BASE_URL
@@ -242,7 +242,7 @@ class DeepSeekSmallModelProvider(BaseServiceProvider[BaseChatModel]):
     """
     
     def __init__(self, model: str = None):
-        from app.config import SMALL_DEEPSEEK_MODEL, SMALL_DEEPSEEK_API_KEY, SMALL_DEEPSEEK_API_BASE
+        from ..config import SMALL_DEEPSEEK_MODEL, SMALL_DEEPSEEK_API_KEY, SMALL_DEEPSEEK_API_BASE
         super().__init__("deepseek_small")
         self._model = model or SMALL_DEEPSEEK_MODEL
         self._api_key = SMALL_DEEPSEEK_API_KEY

backend/app/model_services/embedding_services.py

@@ -21,7 +21,7 @@ from .base import (
     FallbackServiceChain,
     SingletonServiceManager
 )
-from app.config import (
+from ..config import (
     LLAMACPP_EMBEDDING_URL,
     LLAMACPP_API_KEY,
     ZHIPUAI_API_KEY,

backend/app/model_services/rerank_services.py

@@ -27,7 +27,7 @@ from .base import (
     FallbackServiceChain,
     SingletonServiceManager
 )
-from app.config import (
+from ..config import (
     LLAMACPP_RERANKER_URL,
     LLAMACPP_API_KEY,
     ZHIPUAI_API_KEY,
@@ -92,20 +92,33 @@ class LocalLlamaCppRerankService(BaseRerankService):
             "documents": documents,
         }
 
+        logger.info(f"[LocalLlamaCppRerank] 调用 rerank API: {base}/rerank")
+        logger.info(f"[LocalLlamaCppRerank] 请求 payload: query={query[:50]}, documents数量={len(documents)}")
+
         with httpx.Client(timeout=120) as client:
             response = client.post(
                 f"{base}/rerank",
                 headers=headers,
                 json=payload,
             )
-            response.raise_for_status()
+            logger.info(f"[LocalLlamaCppRerank] 响应状态码: {response.status_code}")
+
+            if response.status_code != 200:
+                logger.error(f"[LocalLlamaCppRerank] 请求失败: {response.status_code}")
+                logger.error(f"[LocalLlamaCppRerank] 响应内容: {response.text[:500]}")
+                response.raise_for_status()
+
             data = response.json()
+            logger.info(f"[LocalLlamaCppRerank] 响应数据类型: {type(data)}")
 
             if isinstance(data, dict) and "results" in data:
                 results = data["results"]
                 results_sorted = sorted(results, key=lambda x: x["index"])
-                return [item["relevance_score"] for item in results_sorted]
+                scores = [item["relevance_score"] for item in results_sorted]
+                logger.info(f"[LocalLlamaCppRerank] 返回 {len(scores)} 个得分")
+                return scores
             else:
+                logger.error(f"[LocalLlamaCppRerank] 未知响应格式: {type(data)}")
                 raise ValueError(f"未知的 rerank API 响应格式: {data}")
 
 
@@ -207,11 +220,13 @@ class LLMFallbackRerankService(BaseRerankService):
         if not documents:
             return []
 
+        logger.info(f"[LLMFallbackRerank] 开始为 {len(documents)} 个文档打分")
         scores = []
-        for doc in documents:
+        for i, doc in enumerate(documents):
             score = self._score_single_document(query, doc)
             scores.append(score)
-        
+            logger.info(f"[LLMFallbackRerank] doc[{i}] score={score:.4f}")
+
         return scores
 
     def _score_single_document(self, query: str, document: str) -> float: