refine TGI parmas

app_modules/llm_loader.py

@@ -191,11 +191,11 @@ class LLMLoader:
                 self.llm = HuggingFaceTextGenInference(
                     inference_server_url=HFTGI_SERVER_URL,
                     max_new_tokens=self.max_tokens_limit / 2,
-                    top_k=10,
+                    # top_k=0,
                     top_p=0.95,
-                    typical_p=0.95,
+                    # typical_p=0.95,
                     temperature=0.01,
-                    repetition_penalty=1.03,
+                    repetition_penalty=1.115,
                     callbacks=callbacks,
                     streaming=True,
                 )