diff --git a/src/llmtuner/tuner/core/loader.py b/src/llmtuner/tuner/core/loader.py
index 32d6247a..47c72de1 100644
--- a/src/llmtuner/tuner/core/loader.py
+++ b/src/llmtuner/tuner/core/loader.py
@@ -102,11 +102,11 @@ def load_model_and_tokenizer(
                     )
 
                 current_max_length = getattr(config, "max_position_embeddings", None)
-                if current_max_length and model_args.model_max_length <= current_max_length:
+                if current_max_length and model_args.model_max_length > current_max_length:
+                    scaling_factor = float(math.ceil(model_args.model_max_length / current_max_length))
+                else:
                     logger.warning("Input length is smaller than max length. Consider increase input length.")
                     scaling_factor = 1.0
-                else:
-                    scaling_factor = float(math.ceil(model_args.model_max_length / current_max_length))
             else:
                 scaling_factor = 2.0