diff --git a/flagscale/auto_tuner/tuner.py b/flagscale/auto_tuner/tuner.py
index 66b9223a6..fc48dbf73 100644
--- a/flagscale/auto_tuner/tuner.py
+++ b/flagscale/auto_tuner/tuner.py
@@ -123,7 +123,7 @@ def tune(self):
             # get best strategy
             best_strategy = self.get_best()
             if best_strategy:
-                self.logger.info(f"Best strategy tuned so far: {best_strategy}, and performance is {best_strategy["performance"]}.")
+                self.logger.info(f"Best strategy tuned so far: {best_strategy}, and performance is {best_strategy['performance']}.")
             else:
                 self.logger.info(f"No strategy can run so far.")
         tuner_end_time = time.time()
diff --git a/flagscale/auto_tuner/utils.py b/flagscale/auto_tuner/utils.py
index 3e4409e5c..d1fed7c3a 100644
--- a/flagscale/auto_tuner/utils.py
+++ b/flagscale/auto_tuner/utils.py
@@ -67,13 +67,15 @@ def sort_by_memory(strategy):
         -strategy["use_distributed_optimizer"],
         strategy["micro_batch_size"],
         -strategy["use_recompute"],
+        -strategy["sequence_parallel"],
     )
 
 
 def sort_by_performance(strategy):
     return (
-        -strategy["micro_batch_size"],
-        strategy["use_recompute"],
-        strategy["tensor_model_parallel_size"],
+        -strategy["use_recompute"],
+        (strategy["tensor_model_parallel_size"] % 4),
+        (strategy["micro_batch_size"] % 4),
         strategy["pipeline_model_parallel_size"],
+        -strategy["sequence_parallel"],
     )