diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
index 905744a64ed4c6..0b2a88cb803dad 100755
--- a/src/transformers/trainer.py
+++ b/src/transformers/trainer.py
@@ -1734,6 +1734,15 @@ def _inner_training_loop(
                 )
             elif is_sagemaker_mp_enabled() or self.is_fsdp_enabled:
                 self._load_from_checkpoint(resume_from_checkpoint, self.model_wrapped)
+        
+        # deepspeed ckpt loading
+        if resume_from_checkpoint is not None and self.is_deepspeed_enabled:
+            deepspeed_load_checkpoint(self.model_wrapped, resume_from_checkpoint, load_module_strict=False)
+            if self.args.deepspeed_force_lr_scheduler_checkpointing and self.model_wrapped.lr_scheduler is None:
+                if os.path.isfile(os.path.join(resume_from_checkpoint, SCHEDULER_NAME)):
+                    with warnings.catch_warnings(record=True) as caught_warnings:
+                        self.lr_scheduler.load_state_dict(torch.load(os.path.join(resume_from_checkpoint, SCHEDULER_NAME)))
+                    reissue_pt_warnings(caught_warnings)
 
         # Check if saved optimizer or scheduler states exist
         self._load_optimizer_and_scheduler(resume_from_checkpoint)
@@ -2416,6 +2425,12 @@ def _save_checkpoint(self, model, trial, metrics=None):
         else:
             staging_output_dir = os.path.join(run_dir, f"tmp-{checkpoint_folder}")
         self.save_model(staging_output_dir, _internal_call=True)
+        if self.is_deepspeed_enabled:
+            # under zero3 model file itself doesn't get saved since it's bogus! Unless deepspeed
+            # config `stage3_gather_16bit_weights_on_model_save` is True
+            self.model_wrapped.save_checkpoint(staging_output_dir)
+            if self.args.deepspeed_force_lr_scheduler_checkpointing and self.model_wrapped.lr_scheduler is None:
+                torch.save(self.lr_scheduler.state_dict(), os.path.join(staging_output_dir, SCHEDULER_NAME))
 
         if not self.args.save_only_model:
             # Save optimizer and scheduler
diff --git a/src/transformers/training_args.py b/src/transformers/training_args.py
index 56f102396e0fe5..47aa4d11b472d6 100644
--- a/src/transformers/training_args.py
+++ b/src/transformers/training_args.py
@@ -1316,6 +1316,18 @@ class TrainingArguments:
             "help": "Activates neftune noise embeddings into the model. NEFTune has been proven to drastically improve model performances for instrcution fine-tuning. Check out the original paper here: https://arxiv.org/abs/2310.05914 and the original code here: https://github.com/neelsjain/NEFTune. Only supported for `PreTrainedModel` and `PeftModel` classes."
         },
     )
+    
+    deepspeed_force_lr_scheduler_checkpointing: bool = field(
+        default=False,
+        metadata={
+            "help": (
+                "Force saving and loading or checkpointing the lr_scheduler when deepspeed is enabled and it does not "
+                "support the lr_scheduler type. "
+                "Use this to force keeping track of lr_scheduler when the model lr_scheduler type does not fall into "
+                "its supported lr_scheduler categories."
+            )
+        },
+    )
 
     def __post_init__(self):
         # expand paths, if not os.makedirs("~/bar") will make directory