diff --git a/deepmd/pt/train/training.py b/deepmd/pt/train/training.py
index 62bc5a4c97..fb28f0c4f2 100644
--- a/deepmd/pt/train/training.py
+++ b/deepmd/pt/train/training.py
@@ -132,6 +132,7 @@ def __init__(
         self.disp_freq = training_params.get("disp_freq", 1000)
         self.save_ckpt = training_params.get("save_ckpt", "model.ckpt")
         self.save_freq = training_params.get("save_freq", 1000)
+        self.max_ckpt_keep = training_params.get("max_ckpt_keep", 5)
         self.lcurve_should_print_header = True
 
         def get_opt_param(params):
@@ -924,6 +925,15 @@ def save_model(self, save_path, lr=0.0, step=0):
             {"model": module.state_dict(), "optimizer": self.optimizer.state_dict()},
             save_path,
         )
+        checkpoint_dir = save_path.parent
+        checkpoint_files = [
+            f
+            for f in checkpoint_dir.glob("*.pt")
+            if not f.is_symlink() and f.name.startswith(self.save_ckpt)
+        ]
+        if len(checkpoint_files) > self.max_ckpt_keep:
+            checkpoint_files.sort(key=lambda x: x.stat().st_mtime)
+            checkpoint_files[0].unlink()
 
     def get_data(self, is_train=True, task_key="Default"):
         if not self.multi_task:
diff --git a/deepmd/tf/train/trainer.py b/deepmd/tf/train/trainer.py
index 1dd31fd0bb..27478abaa1 100644
--- a/deepmd/tf/train/trainer.py
+++ b/deepmd/tf/train/trainer.py
@@ -164,6 +164,7 @@ def get_lr_and_coef(lr_param):
         self.disp_freq = tr_data.get("disp_freq", 1000)
         self.save_freq = tr_data.get("save_freq", 1000)
         self.save_ckpt = tr_data.get("save_ckpt", "model.ckpt")
+        self.max_ckpt_keep = tr_data.get("max_ckpt_keep", 5)
         self.display_in_training = tr_data.get("disp_training", True)
         self.timing_in_training = tr_data.get("time_training", True)
         self.profiling = self.run_opt.is_chief and tr_data.get("profiling", False)
@@ -498,7 +499,9 @@ def _init_session(self):
         # Initializes or restore global variables
         init_op = tf.global_variables_initializer()
         if self.run_opt.is_chief:
-            self.saver = tf.train.Saver(save_relative_paths=True)
+            self.saver = tf.train.Saver(
+                save_relative_paths=True, max_to_keep=self.max_ckpt_keep
+            )
             if self.run_opt.init_mode == "init_from_scratch":
                 log.info("initialize model from scratch")
                 run_sess(self.sess, init_op)
diff --git a/deepmd/utils/argcheck.py b/deepmd/utils/argcheck.py
index 5e8db431f8..e822e18d50 100644
--- a/deepmd/utils/argcheck.py
+++ b/deepmd/utils/argcheck.py
@@ -2134,6 +2134,11 @@ def training_args():  # ! modified by Ziyao: data configuration isolated.
     doc_disp_freq = "The frequency of printing learning curve."
     doc_save_freq = "The frequency of saving check point."
     doc_save_ckpt = "The path prefix of saving check point files."
+    doc_max_ckpt_keep = (
+        "The maximum number of checkpoints to keep. "
+        "The oldest checkpoints will be deleted once the number of checkpoints exceeds max_ckpt_keep. "
+        "Defaults to 5."
+    )
     doc_disp_training = "Displaying verbose information during training."
     doc_time_training = "Timing durining training."
     doc_profiling = "Profiling during training."
@@ -2192,6 +2197,7 @@ def training_args():  # ! modified by Ziyao: data configuration isolated.
         Argument(
             "save_ckpt", str, optional=True, default="model.ckpt", doc=doc_save_ckpt
         ),
+        Argument("max_ckpt_keep", int, optional=True, default=5, doc=doc_max_ckpt_keep),
         Argument(
             "disp_training", bool, optional=True, default=True, doc=doc_disp_training
         ),