fix FSDP model resume optimizer & scheduler (#25852)

* fix FSDP resume optimizer & scheduler

* improve trainer code quality

---------
Co-authored-by: machi04 <machi04@meituan.com>

src/transformers/trainer.py

@@ -223,6 +223,7 @@ logger = logging.get_logger(__name__)
 TRAINING_ARGS_NAME = "training_args.bin"
 TRAINER_STATE_NAME = "trainer_state.json"
 OPTIMIZER_NAME = "optimizer.pt"
+OPTIMIZER_NAME_BIN = "optimizer.bin"
 SCHEDULER_NAME = "scheduler.pt"
 SCALER_NAME = "scaler.pt"
 
@@ -2360,16 +2361,12 @@ class Trainer:
                     partial=True,
                     v3=smp.state.cfg.shard_optimizer_state,
                 )
-            if self.args.should_save:
-                with warnings.catch_warnings(record=True) as caught_warnings:
-                    torch.save(self.lr_scheduler.state_dict(), os.path.join(output_dir, SCHEDULER_NAME))
-                reissue_pt_warnings(caught_warnings)
-                if self.do_grad_scaling:
-                    torch.save(self.scaler.state_dict(), os.path.join(output_dir, SCALER_NAME))
         elif self.args.should_save and not self.is_deepspeed_enabled and not (self.fsdp or self.is_fsdp_enabled):
             # deepspeed.save_checkpoint above saves model/optim/sched
             torch.save(self.optimizer.state_dict(), os.path.join(output_dir, OPTIMIZER_NAME))
 
+        # Save SCHEDULER & SCALER
+        if self.args.should_save and not self.is_deepspeed_enabled and not is_torch_tpu_available():
             with warnings.catch_warnings(record=True) as caught_warnings:
                 torch.save(self.lr_scheduler.state_dict(), os.path.join(output_dir, SCHEDULER_NAME))
             reissue_pt_warnings(caught_warnings)
@@ -2440,7 +2437,10 @@ class Trainer:
         checkpoint_file_exists = (
             glob.glob(os.path.join(checkpoint, OPTIMIZER_NAME) + "_*")
             if is_sagemaker_mp_enabled()
-            else os.path.isfile(os.path.join(checkpoint, OPTIMIZER_NAME))
+            else (
+                os.path.isfile(os.path.join(checkpoint, OPTIMIZER_NAME))
+                or os.path.isfile(os.path.join(checkpoint, OPTIMIZER_NAME_BIN))
+            )
         )
         if checkpoint_file_exists and os.path.isfile(os.path.join(checkpoint, SCHEDULER_NAME)):
             # Load in optimizer and scheduler states