diff --git a/api/apps/llm_app.py b/api/apps/llm_app.py
index 5e045e1c396..056dba8c7ca 100644
--- a/api/apps/llm_app.py
+++ b/api/apps/llm_app.py
@@ -43,7 +43,7 @@ def set_api_key():
     chat_passed, embd_passed, rerank_passed = False, False, False
     factory = req["llm_factory"]
     msg = ""
-    for llm in LLMService.query(fid=factory):
+    for llm in LLMService.query(fid=factory)[:3]:
         if not embd_passed and llm.model_type == LLMType.EMBEDDING.value:
             mdl = EmbeddingModel[factory](
                 req["api_key"], llm.llm_name, base_url=req.get("base_url"))
diff --git a/rag/llm/sequence2txt_model.py b/rag/llm/sequence2txt_model.py
index 626a65e452b..fcd6859d598 100644
--- a/rag/llm/sequence2txt_model.py
+++ b/rag/llm/sequence2txt_model.py
@@ -71,7 +71,7 @@ def transcription(self, audio, format):
         ans = ""
         if result.status_code == HTTPStatus.OK:
             for sentence in result.get_sentence():
-                ans += str(sentence + '\n')
+                ans += sentence.text.decode('utf-8') + '\n'
             return ans, num_tokens_from_string(ans)
 
         return "**ERROR**: " + result.message, 0