diff --git a/api/db/services/dialog_service.py b/api/db/services/dialog_service.py
index fb9ed5bd0..0a4ab0f0e 100644
--- a/api/db/services/dialog_service.py
+++ b/api/db/services/dialog_service.py
@@ -859,7 +859,7 @@ Output:
 
 
 def reasoning(chunk_info: dict, question: str, chat_mdl: LLMBundle, embd_mdl: LLMBundle,
-              tenant_ids: list[str], kb_ids: list[str], prompt_config, MAX_SEARCH_LIMIT: int = 3,
+              tenant_ids: list[str], kb_ids: list[str], prompt_config, MAX_SEARCH_LIMIT: int = 6,
               top_n: int = 5, similarity_threshold: float = 0.4, vector_similarity_weight: float = 0.3):
     BEGIN_SEARCH_QUERY = "<|begin_search_query|>"
     END_SEARCH_QUERY = "<|end_search_query|>"
diff --git a/rag/llm/rerank_model.py b/rag/llm/rerank_model.py
index 6caa28b0c..443075153 100644
--- a/rag/llm/rerank_model.py
+++ b/rag/llm/rerank_model.py
@@ -130,12 +130,12 @@ class DefaultRerank(Base):
 
         self._dynamic_batch_size = old_dynamic_batch_size
         return np.array(res)
-    
 
     def _compute_batch_scores(self, batch_pairs, max_length=None):
         if max_length is None:
-            max_length = self._model.max_length
-        scores = self._model.compute_score(batch_pairs, max_length=max_length)
+            scores = self._model.compute_score(batch_pairs)
+        else:
+            scores = self._model.compute_score(batch_pairs, max_length=max_length)
         scores = sigmoid(np.array(scores)).tolist()
         return scores