diff --git a/FlagEmbedding/abc/finetune/embedder/AbsRunner.py b/FlagEmbedding/abc/finetune/embedder/AbsRunner.py
index 9466a3db..0ad49493 100644
--- a/FlagEmbedding/abc/finetune/embedder/AbsRunner.py
+++ b/FlagEmbedding/abc/finetune/embedder/AbsRunner.py
@@ -43,10 +43,11 @@ def __init__(
             os.path.exists(training_args.output_dir)
             and os.listdir(training_args.output_dir)
             and training_args.do_train
-            and not training_args.overwrite_output_dir
+            and training_args.resume_from_checkpoint is None
         ):
             raise ValueError(
-                f"Output directory ({training_args.output_dir}) already exists and is not empty. Use --overwrite_output_dir to overcome."
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Please use a new output directory or set --resume_from_checkpoint."
             )
 
         # Setup logging
diff --git a/FlagEmbedding/abc/finetune/embedder/AbsTrainer.py b/FlagEmbedding/abc/finetune/embedder/AbsTrainer.py
index 38944ed8..0f726c7c 100644
--- a/FlagEmbedding/abc/finetune/embedder/AbsTrainer.py
+++ b/FlagEmbedding/abc/finetune/embedder/AbsTrainer.py
@@ -1,4 +1,5 @@
 import logging
+import inspect
 from typing import Optional
 from abc import ABC, abstractmethod
 from transformers.trainer import Trainer
@@ -10,6 +11,39 @@ class AbsEmbedderTrainer(ABC, Trainer):
     """
     Abstract class for the trainer of embedder.
     """
+    def __init__(
+        self,
+        *args,
+        processing_class=None,
+        tokenizer=None,
+        **kwargs,
+    ):
+        """Initialize ``Trainer`` across Transformers API versions.
+
+        Transformers 5 renamed the ``tokenizer`` argument to
+        ``processing_class``.  FlagEmbedding accepts both names so callers
+        from older examples remain compatible with either Transformers API.
+        """
+        if processing_class is None:
+            processing_class = tokenizer
+
+        trainer_parameters = inspect.signature(Trainer.__init__).parameters
+        if "processing_class" in trainer_parameters:
+            kwargs["processing_class"] = processing_class
+        else:
+            kwargs["tokenizer"] = processing_class
+
+        super().__init__(*args, **kwargs)
+
+    def _save_processing_class(self, output_dir: str):
+        """Save the tokenizer/processor using the active Transformers API."""
+        processing_class = getattr(self, "processing_class", None)
+        if processing_class is None:
+            processing_class = getattr(self, "tokenizer", None)
+
+        if processing_class is not None and self.is_world_process_zero():
+            processing_class.save_pretrained(output_dir)
+
     @abstractmethod
     def _save(self, output_dir: Optional[str] = None, state_dict=None):
         pass
diff --git a/FlagEmbedding/abc/finetune/reranker/AbsRunner.py b/FlagEmbedding/abc/finetune/reranker/AbsRunner.py
index b40c2d2a..875776c1 100644
--- a/FlagEmbedding/abc/finetune/reranker/AbsRunner.py
+++ b/FlagEmbedding/abc/finetune/reranker/AbsRunner.py
@@ -43,10 +43,11 @@ def __init__(
             os.path.exists(training_args.output_dir)
             and os.listdir(training_args.output_dir)
             and training_args.do_train
-            and not training_args.overwrite_output_dir
+            and training_args.resume_from_checkpoint is None
         ):
             raise ValueError(
-                f"Output directory ({training_args.output_dir}) already exists and is not empty. Use --overwrite_output_dir to overcome."
+                f"Output directory ({training_args.output_dir}) already exists and is not empty. "
+                "Please use a new output directory or set --resume_from_checkpoint."
             )
 
         # Setup logging
diff --git a/FlagEmbedding/abc/finetune/reranker/AbsTrainer.py b/FlagEmbedding/abc/finetune/reranker/AbsTrainer.py
index 213d5f79..afd11b6e 100644
--- a/FlagEmbedding/abc/finetune/reranker/AbsTrainer.py
+++ b/FlagEmbedding/abc/finetune/reranker/AbsTrainer.py
@@ -1,4 +1,5 @@
 import logging
+import inspect
 from typing import Optional
 from abc import ABC, abstractmethod
 from transformers.trainer import Trainer
@@ -10,6 +11,34 @@ class AbsRerankerTrainer(ABC, Trainer):
     """
     Abstract class for the trainer of reranker.
     """
+    def __init__(
+        self,
+        *args,
+        processing_class=None,
+        tokenizer=None,
+        **kwargs,
+    ):
+        """Initialize ``Trainer`` across Transformers API versions."""
+        if processing_class is None:
+            processing_class = tokenizer
+
+        trainer_parameters = inspect.signature(Trainer.__init__).parameters
+        if "processing_class" in trainer_parameters:
+            kwargs["processing_class"] = processing_class
+        else:
+            kwargs["tokenizer"] = processing_class
+
+        super().__init__(*args, **kwargs)
+
+    def _save_processing_class(self, output_dir: str):
+        """Save the tokenizer/processor using the active Transformers API."""
+        processing_class = getattr(self, "processing_class", None)
+        if processing_class is None:
+            processing_class = getattr(self, "tokenizer", None)
+
+        if processing_class is not None and self.is_world_process_zero():
+            processing_class.save_pretrained(output_dir)
+
     @abstractmethod
     def _save(self, output_dir: Optional[str] = None, state_dict=None):
         pass
diff --git a/FlagEmbedding/finetune/embedder/decoder_only/base/trainer.py b/FlagEmbedding/finetune/embedder/decoder_only/base/trainer.py
index bb834c66..36122a57 100644
--- a/FlagEmbedding/finetune/embedder/decoder_only/base/trainer.py
+++ b/FlagEmbedding/finetune/embedder/decoder_only/base/trainer.py
@@ -33,8 +33,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
         else:
             self.model.save(output_dir)
 
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
 
diff --git a/FlagEmbedding/finetune/embedder/decoder_only/icl/trainer.py b/FlagEmbedding/finetune/embedder/decoder_only/icl/trainer.py
index c5d166f9..ee2224f3 100644
--- a/FlagEmbedding/finetune/embedder/decoder_only/icl/trainer.py
+++ b/FlagEmbedding/finetune/embedder/decoder_only/icl/trainer.py
@@ -33,8 +33,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
         else:
             self.model.save(output_dir)
 
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
 
diff --git a/FlagEmbedding/finetune/embedder/encoder_only/base/trainer.py b/FlagEmbedding/finetune/embedder/encoder_only/base/trainer.py
index affa34a8..46d674ba 100644
--- a/FlagEmbedding/finetune/embedder/encoder_only/base/trainer.py
+++ b/FlagEmbedding/finetune/embedder/encoder_only/base/trainer.py
@@ -32,8 +32,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
                 f'does not support save interface')
         else:
             self.model.save(output_dir)
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
 
diff --git a/FlagEmbedding/finetune/embedder/encoder_only/m3/trainer.py b/FlagEmbedding/finetune/embedder/encoder_only/m3/trainer.py
index a6bf5dd0..4aa179ec 100644
--- a/FlagEmbedding/finetune/embedder/encoder_only/m3/trainer.py
+++ b/FlagEmbedding/finetune/embedder/encoder_only/m3/trainer.py
@@ -32,8 +32,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
                 f'does not support save interface')
         else:
             self.model.save(output_dir)
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
 
diff --git a/FlagEmbedding/finetune/reranker/decoder_only/base/runner.py b/FlagEmbedding/finetune/reranker/decoder_only/base/runner.py
index 14fa3dd9..e2e4c457 100644
--- a/FlagEmbedding/finetune/reranker/decoder_only/base/runner.py
+++ b/FlagEmbedding/finetune/reranker/decoder_only/base/runner.py
@@ -89,7 +89,7 @@ def load_trainer(self) -> DecoderOnlyRerankerTrainer:
             args=self.training_args,
             train_dataset=self.train_dataset,
             data_collator=self.data_collator,
-            tokenizer=self.tokenizer
+            processing_class=self.tokenizer
         )
         return trainer
 
diff --git a/FlagEmbedding/finetune/reranker/decoder_only/base/trainer.py b/FlagEmbedding/finetune/reranker/decoder_only/base/trainer.py
index d74ad0b5..2331df85 100644
--- a/FlagEmbedding/finetune/reranker/decoder_only/base/trainer.py
+++ b/FlagEmbedding/finetune/reranker/decoder_only/base/trainer.py
@@ -35,8 +35,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
         else:
             self.model.save(output_dir)
             
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
 
diff --git a/FlagEmbedding/finetune/reranker/decoder_only/layerwise/runner.py b/FlagEmbedding/finetune/reranker/decoder_only/layerwise/runner.py
index b2502004..79786c82 100644
--- a/FlagEmbedding/finetune/reranker/decoder_only/layerwise/runner.py
+++ b/FlagEmbedding/finetune/reranker/decoder_only/layerwise/runner.py
@@ -90,7 +90,7 @@ def load_trainer(self) -> DecoderOnlyRerankerTrainer:
             args=self.training_args,
             train_dataset=self.train_dataset,
             data_collator=self.data_collator,
-            tokenizer=self.tokenizer
+            processing_class=self.tokenizer
         )
         return trainer
 
diff --git a/FlagEmbedding/finetune/reranker/decoder_only/layerwise/trainer.py b/FlagEmbedding/finetune/reranker/decoder_only/layerwise/trainer.py
index dc8c7a18..2f71682d 100644
--- a/FlagEmbedding/finetune/reranker/decoder_only/layerwise/trainer.py
+++ b/FlagEmbedding/finetune/reranker/decoder_only/layerwise/trainer.py
@@ -35,8 +35,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
         else:
             self.model.save(output_dir)
 
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
 
diff --git a/FlagEmbedding/finetune/reranker/encoder_only/base/runner.py b/FlagEmbedding/finetune/reranker/encoder_only/base/runner.py
index 324c9e30..7842ee0b 100644
--- a/FlagEmbedding/finetune/reranker/encoder_only/base/runner.py
+++ b/FlagEmbedding/finetune/reranker/encoder_only/base/runner.py
@@ -71,6 +71,6 @@ def load_trainer(self) -> EncoderOnlyRerankerTrainer:
             args=self.training_args,
             train_dataset=self.train_dataset,
             data_collator=self.data_collator,
-            tokenizer=self.tokenizer
+            processing_class=self.tokenizer
         )
         return trainer
diff --git a/FlagEmbedding/finetune/reranker/encoder_only/base/trainer.py b/FlagEmbedding/finetune/reranker/encoder_only/base/trainer.py
index 59120341..36ac40ce 100644
--- a/FlagEmbedding/finetune/reranker/encoder_only/base/trainer.py
+++ b/FlagEmbedding/finetune/reranker/encoder_only/base/trainer.py
@@ -30,8 +30,7 @@ def _save(self, output_dir: Optional[str] = None, state_dict=None):
             raise NotImplementedError(f'MODEL {self.model.__class__.__name__} ' f'does not support save_pretrained interface')
         else:
             self.model.save_pretrained(output_dir)
-        if self.tokenizer is not None and self.is_world_process_zero():
-            self.tokenizer.save_pretrained(output_dir)
+        self._save_processing_class(output_dir)
 
         # Good practice: save your training arguments together with the trained model
         torch.save(self.args, os.path.join(output_dir, "training_args.bin"))
diff --git a/examples/README.md b/examples/README.md
index 3faea48d..abb6d8c8 100644
--- a/examples/README.md
+++ b/examples/README.md
@@ -104,13 +104,11 @@ torchrun --nproc_per_node 2 \
     --query_instruction_format '{}{}' \
     --knowledge_distillation False \
     --output_dir ./test_encoder_only_base_bge-large-en-v1.5 \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 2 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ./finetune/ds_stage0.json \
     --logging_steps 1 \
@@ -120,6 +118,7 @@ torchrun --nproc_per_node 2 \
     --sentence_pooling_method cls \
     --normalize_embeddings True \
     --kd_loss_type kl_div
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 ### 2. Reranker
@@ -137,19 +136,18 @@ torchrun --nproc_per_node 2 \
     --pad_to_multiple_of 8 \
     --knowledge_distillation False \
     --output_dir ./test_encoder_only_base_bge-reranker-large \
-    --overwrite_output_dir \
     --learning_rate 6e-5 \
     --fp16 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 2 \
     --gradient_accumulation_steps 1 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ./finetune/ds_stage0.json \
     --logging_steps 1 \
     --save_steps 1000
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 ## 5. Evaluation
@@ -187,4 +185,3 @@ python -m FlagEmbedding.evaluation.msmarco \
     --devices cuda:0 cuda:1 cuda:2 cuda:3 cuda:4 cuda:5 cuda:6 cuda:7 \
     --cache_dir ./cache/model
 ```
-
diff --git a/examples/finetune/embedder/README.md b/examples/finetune/embedder/README.md
index c9e141aa..b1197ee5 100644
--- a/examples/finetune/embedder/README.md
+++ b/examples/finetune/embedder/README.md
@@ -184,13 +184,11 @@ torchrun --nproc_per_node 2 \
     --query_instruction_format '{}{}' \
     --knowledge_distillation False \
 	--output_dir ./test_encoder_only_base_bge-large-en-v1.5 \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs 2 \
     --per_device_train_batch_size 2 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../ds_stage0.json \
     --logging_steps 1 \
@@ -200,6 +198,7 @@ torchrun --nproc_per_node 2 \
     --sentence_pooling_method cls \
     --normalize_embeddings True \
     --kd_loss_type kl_div
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 ### (2) bge-m3
@@ -223,13 +222,11 @@ torchrun --nproc_per_node 2 \
     --small_threshold 0 \
     --drop_threshold 0 \
     --output_dir ./test_encoder_only_m3_bge-m3_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs 2 \
     --per_device_train_batch_size 2 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../ds_stage0.json \
     --logging_steps 1 \
@@ -243,6 +240,7 @@ torchrun --nproc_per_node 2 \
     --use_self_distill True \
     --fix_encoder False \
     --self_distill_start_step 0
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 Here are some new arguments:
@@ -282,13 +280,11 @@ torchrun --nproc_per_node 2 \
     --small_threshold 0 \
     --drop_threshold 0 \
 	--output_dir ./test_decoder_only_base_bge-multilingual-gemma2_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-4 \
     --fp16 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 2 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../ds_stage1.json \
     --logging_steps 1 \
@@ -298,6 +294,7 @@ torchrun --nproc_per_node 2 \
     --sentence_pooling_method last_token \
     --normalize_embeddings True \
     --kd_loss_type m3_kd_loss
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 Here are some new arguments:
@@ -346,13 +343,11 @@ torchrun --nproc_per_node 2 \
     --retrieval_use_examples True \
     --icl_suffix_str '\n<response>' \
     --output_dir ./test_decoder_only_base_bge-en-icl_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-4 \
     --fp16 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 2 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../ds_stage1.json \
     --logging_steps 1 \
@@ -362,6 +357,7 @@ torchrun --nproc_per_node 2 \
     --sentence_pooling_method last_token \
     --normalize_embeddings True \
     --kd_loss_type kl_div
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 Here are some new arguments:
@@ -383,4 +379,3 @@ Here are some new arguments:
 - **`example_passage_max_len`**: The max length of example passage.
 - **`retrieval_use_examples`**: If passed, will use examples for retrieval.
 - **`icl_suffix_str`**: The suffix string for ICL dataset.
-
diff --git a/examples/finetune/embedder/decoder_only/base.sh b/examples/finetune/embedder/decoder_only/base.sh
index 2b91f683..8268a222 100644
--- a/examples/finetune/embedder/decoder_only/base.sh
+++ b/examples/finetune/embedder/decoder_only/base.sh
@@ -42,13 +42,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_decoder_only_base_bge-multilingual-gemma2 \
-    --overwrite_output_dir \
     --learning_rate 1e-4 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage1.json \
     --logging_steps 1 \
@@ -60,6 +58,8 @@ training_args="\
     --kd_loss_type m3_kd_loss \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.decoder_only.base \
     $model_args \
diff --git a/examples/finetune/embedder/decoder_only/base_same_dataset.sh b/examples/finetune/embedder/decoder_only/base_same_dataset.sh
index d002264b..197e5288 100644
--- a/examples/finetune/embedder/decoder_only/base_same_dataset.sh
+++ b/examples/finetune/embedder/decoder_only/base_same_dataset.sh
@@ -45,13 +45,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_decoder_only_base_bge-multilingual-gemma2_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-4 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage1.json \
     --logging_steps 1 \
@@ -63,6 +61,8 @@ training_args="\
     --kd_loss_type m3_kd_loss \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.decoder_only.base \
     $model_args \
diff --git a/examples/finetune/embedder/decoder_only/icl_same_dataset.sh b/examples/finetune/embedder/decoder_only/icl_same_dataset.sh
index 025ece9e..2e080f04 100644
--- a/examples/finetune/embedder/decoder_only/icl_same_dataset.sh
+++ b/examples/finetune/embedder/decoder_only/icl_same_dataset.sh
@@ -49,13 +49,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_decoder_only_base_bge-en-icl_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-4 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage1.json \
     --logging_steps 1 \
@@ -67,6 +65,8 @@ training_args="\
     --kd_loss_type kl_div \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.decoder_only.icl \
     $model_args \
diff --git a/examples/finetune/embedder/encoder_only/base.sh b/examples/finetune/embedder/encoder_only/base.sh
index aae9e682..c345fa8d 100644
--- a/examples/finetune/embedder/encoder_only/base.sh
+++ b/examples/finetune/embedder/encoder_only/base.sh
@@ -36,13 +36,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_encoder_only_base_bge-large-en-v1.5 \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage0.json \
     --logging_steps 1 \
@@ -54,6 +52,8 @@ training_args="\
     --kd_loss_type kl_div \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.encoder_only.base \
     $model_args \
diff --git a/examples/finetune/embedder/encoder_only/base_same_dataset.sh b/examples/finetune/embedder/encoder_only/base_same_dataset.sh
index fe9a6b31..944e463e 100644
--- a/examples/finetune/embedder/encoder_only/base_same_dataset.sh
+++ b/examples/finetune/embedder/encoder_only/base_same_dataset.sh
@@ -39,13 +39,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_encoder_only_base_bge-large-en-v1.5_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage0.json \
     --logging_steps 1 \
@@ -57,6 +55,8 @@ training_args="\
     --kd_loss_type kl_div \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.encoder_only.base \
     $model_args \
diff --git a/examples/finetune/embedder/encoder_only/m3.sh b/examples/finetune/embedder/encoder_only/m3.sh
index 49642a40..201cebb8 100644
--- a/examples/finetune/embedder/encoder_only/m3.sh
+++ b/examples/finetune/embedder/encoder_only/m3.sh
@@ -34,13 +34,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_encoder_only_m3_bge-m3 \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage0.json \
     --logging_steps 1 \
@@ -56,6 +54,8 @@ training_args="\
     --self_distill_start_step 0 \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.encoder_only.m3 \
     $model_args \
diff --git a/examples/finetune/embedder/encoder_only/m3_same_dataset.sh b/examples/finetune/embedder/encoder_only/m3_same_dataset.sh
index e0b7c8d8..3b7d1296 100644
--- a/examples/finetune/embedder/encoder_only/m3_same_dataset.sh
+++ b/examples/finetune/embedder/encoder_only/m3_same_dataset.sh
@@ -37,13 +37,11 @@ data_args="\
 
 training_args="\
     --output_dir ./test_encoder_only_m3_bge-m3_sd \
-    --overwrite_output_dir \
     --learning_rate 1e-5 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --deepspeed ../../ds_stage0.json \
     --logging_steps 1 \
@@ -59,6 +57,8 @@ training_args="\
     --self_distill_start_step 0 \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.embedder.encoder_only.m3 \
     $model_args \
diff --git a/examples/finetune/reranker/README.md b/examples/finetune/reranker/README.md
index 1e53ddf2..4fde383d 100644
--- a/examples/finetune/reranker/README.md
+++ b/examples/finetune/reranker/README.md
@@ -158,19 +158,18 @@ torchrun --nproc_per_node 2 \
     --pad_to_multiple_of 8 \
     --knowledge_distillation False \
 	--output_dir ./test_encoder_only_base_bge-reranker-base \
-    --overwrite_output_dir \
     --learning_rate 6e-5 \
     --fp16 \
     --num_train_epochs 2 \
     --per_device_train_batch_size 2 \
     --gradient_accumulation_steps 1 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ../ds_stage0.json \
     --logging_steps 1 \
     --save_steps 1000
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 ### (2) bge-reranker-v2-gemma
@@ -199,19 +198,18 @@ torchrun --nproc_per_node 2 \
     --passage_instruction_for_rerank 'B: ' \
     --passage_instruction_format '{}{}' \
     --output_dir ./test_decoder_only_base_bge-reranker-v2-minicpm-layerwise \
-    --overwrite_output_dir \
     --learning_rate 2e-4 \
     --bf16 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 2 \
     --gradient_accumulation_steps 1 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ../ds_stage0.json \
     --logging_steps 1 \
     --save_steps 1000
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 Here are some new arguments:
@@ -258,19 +256,18 @@ torchrun --nproc_per_node 2 \
     --passage_instruction_for_rerank 'B: ' \
     --passage_instruction_format '{}{}' \
 	--output_dir ./test_decoder_only_base_bge-reranker-v2-minicpm-layerwise \
-    --overwrite_output_dir \
     --learning_rate 2e-4 \
     --bf16 \
     --num_train_epochs 1 \
     --per_device_train_batch_size 2 \
     --gradient_accumulation_steps 1 \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ../ds_stage0.json \
     --logging_steps 1 \
     --save_steps 1000
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
 ```
 
 Here are some new arguments:
diff --git a/examples/finetune/reranker/decoder_only/base.sh b/examples/finetune/reranker/decoder_only/base.sh
index 1f862a88..2230edfa 100644
--- a/examples/finetune/reranker/decoder_only/base.sh
+++ b/examples/finetune/reranker/decoder_only/base.sh
@@ -44,14 +44,12 @@ data_args="\
 
 training_args="\
     --output_dir ./test_decoder_only_base_bge-reranker-v2-gemma \
-    --overwrite_output_dir \
     --learning_rate 2e-4 \
     --bf16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --gradient_accumulation_steps $gradient_accumulation_steps \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ../../ds_stage0.json \
@@ -59,6 +57,8 @@ training_args="\
     --save_steps 1000 \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.reranker.decoder_only.base \
     $model_args \
@@ -67,4 +67,4 @@ cmd="torchrun --nproc_per_node $num_gpus \
 "
 
 echo $cmd
-eval $cmd
\ No newline at end of file
+eval $cmd
diff --git a/examples/finetune/reranker/decoder_only/layerwise.sh b/examples/finetune/reranker/decoder_only/layerwise.sh
index f5d44c8b..8d2b6848 100644
--- a/examples/finetune/reranker/decoder_only/layerwise.sh
+++ b/examples/finetune/reranker/decoder_only/layerwise.sh
@@ -49,14 +49,12 @@ data_args="\
 
 training_args="\
     --output_dir ./test_decoder_only_base_bge-reranker-v2-minicpm-layerwise \
-    --overwrite_output_dir \
     --learning_rate 2e-4 \
     --bf16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --gradient_accumulation_steps $gradient_accumulation_steps \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ../../ds_stage0.json \
@@ -64,6 +62,8 @@ training_args="\
     --save_steps 1000 \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.reranker.decoder_only.layerwise \
     $model_args \
@@ -73,4 +73,3 @@ cmd="torchrun --nproc_per_node $num_gpus \
 
 echo $cmd
 eval $cmd
-
diff --git a/examples/finetune/reranker/encoder_only/base.sh b/examples/finetune/reranker/encoder_only/base.sh
index c8ca09fc..d0cbea6e 100644
--- a/examples/finetune/reranker/encoder_only/base.sh
+++ b/examples/finetune/reranker/encoder_only/base.sh
@@ -33,14 +33,12 @@ data_args="\
 
 training_args="\
     --output_dir ./test_encoder_only_base_bge-reranker-base \
-    --overwrite_output_dir \
     --learning_rate 6e-5 \
     --fp16 \
     --num_train_epochs $num_train_epochs \
     --per_device_train_batch_size $per_device_train_batch_size \
     --gradient_accumulation_steps $gradient_accumulation_steps \
     --dataloader_drop_last True \
-    --warmup_ratio 0.1 \
     --gradient_checkpointing \
     --weight_decay 0.01 \
     --deepspeed ../../ds_stage0.json \
@@ -48,6 +46,8 @@ training_args="\
     --save_steps 1000 \
 "
 
+# For transformers<=4.57.3, add --warmup_ratio 0.1; for transformers>=5.0.0, add --warmup_steps 0.1.
+
 cmd="torchrun --nproc_per_node $num_gpus \
     -m FlagEmbedding.finetune.reranker.encoder_only.base \
     $model_args \
@@ -56,4 +56,4 @@ cmd="torchrun --nproc_per_node $num_gpus \
 "
 
 echo $cmd
-eval $cmd
\ No newline at end of file
+eval $cmd
diff --git a/setup.py b/setup.py
index 9e4feacb..cf067d92 100644
--- a/setup.py
+++ b/setup.py
@@ -5,7 +5,7 @@
 
 setup(
     name='FlagEmbedding',
-    version='1.4.0',
+    version='1.4.1',
     description='FlagEmbedding',
     long_description=readme,
     long_description_content_type="text/markdown",
diff --git a/tests/README.md b/tests/README.md
index 36e447c7..79939ae9 100644
--- a/tests/README.md
+++ b/tests/README.md
@@ -1,10 +1,11 @@
 # FlagEmbedding Tests
 
-This directory contains tests for the FlagEmbedding library, including compatibility tests for Transformers 5.0.
+This directory contains tests for the FlagEmbedding library, including compatibility tests for Transformers 5.x.
 
 ## Test Files
 
 - `test_imports_v5.py`: Tests that imports work with Transformers v5, particularly the compatibility layer for `is_torch_fx_available`.
+- `test_finetune_trainer_compat.py`: Tests the Transformers Trainer API migration, including `processing_class`/legacy `tokenizer` construction, reranker runner arguments, and processor checkpoint saving.
 - `test_infer_embedder_basic.py`: Tests basic functionality of BGE embedder models with a small public checkpoint.
 - `test_infer_reranker_basic.py`: Tests basic functionality of reranker models.
 
@@ -24,13 +25,16 @@ pytest tests/
 # Run a specific test file
 pytest tests/test_imports_v5.py
 
+# Run the fine-tuning Trainer compatibility tests
+pytest tests/test_finetune_trainer_compat.py
+
 # Run with verbose output
 pytest -v tests/
 ```
 
-## Transformers 5.0 Compatibility
+## Transformers 5.x Compatibility
 
-The tests verify that FlagEmbedding works with Transformers 5.0, which removed the `is_torch_fx_available` function.
+The tests verify that FlagEmbedding works with Transformers 5.x, which removed the `is_torch_fx_available` function and renamed the Trainer `tokenizer` argument to `processing_class`.
 The compatibility layer in `FlagEmbedding/utils/transformers_compat.py` provides this function for backward compatibility.
 
 **Note:** Transformers 5.0 requires Python 3.10 or higher. If you're using Python 3.9 or lower, you'll need to upgrade your Python version to test with Transformers 5.0.
@@ -39,4 +43,4 @@ To test with a specific version of transformers (with Python 3.10+):
 
 ```bash
 pip install transformers==5.0.0
-pytest tests/
\ No newline at end of file
+pytest tests/
diff --git a/tests/test_finetune_trainer_compat.py b/tests/test_finetune_trainer_compat.py
new file mode 100644
index 00000000..b564590b
--- /dev/null
+++ b/tests/test_finetune_trainer_compat.py
@@ -0,0 +1,153 @@
+"""Regression tests for the Transformers Trainer API migration."""
+
+from pathlib import Path
+
+import torch
+from transformers import Trainer, TrainingArguments
+
+from FlagEmbedding.finetune.embedder.decoder_only.base.trainer import (
+    DecoderOnlyEmbedderTrainer,
+)
+from FlagEmbedding.finetune.embedder.decoder_only.icl.trainer import (
+    DecoderOnlyEmbedderICLTrainer,
+)
+from FlagEmbedding.finetune.embedder.encoder_only.base.trainer import (
+    EncoderOnlyEmbedderTrainer,
+)
+from FlagEmbedding.finetune.embedder.encoder_only.m3.trainer import (
+    EncoderOnlyEmbedderM3Trainer,
+)
+from FlagEmbedding.finetune.reranker.decoder_only.base.trainer import (
+    DecoderOnlyRerankerTrainer,
+)
+from FlagEmbedding.finetune.reranker.decoder_only.base.runner import (
+    DecoderOnlyRerankerRunner,
+)
+from FlagEmbedding.finetune.reranker.decoder_only.layerwise.trainer import (
+    DecoderOnlyRerankerTrainer as DecoderOnlyRerankerLayerwiseTrainer,
+)
+from FlagEmbedding.finetune.reranker.decoder_only.layerwise.runner import (
+    DecoderOnlyRerankerRunner as DecoderOnlyRerankerLayerwiseRunner,
+)
+from FlagEmbedding.finetune.reranker.encoder_only.base.trainer import (
+    EncoderOnlyRerankerTrainer,
+)
+from FlagEmbedding.finetune.reranker.encoder_only.base.runner import (
+    EncoderOnlyRerankerRunner,
+)
+
+
+class RecordingProcessor:
+    def __init__(self, marker: str):
+        self.marker = marker
+
+    def save_pretrained(self, output_dir):
+        Path(output_dir, "processor.marker").write_text(self.marker, encoding="utf-8")
+
+
+class SaveableModel(torch.nn.Module):
+    def save(self, output_dir):
+        Path(output_dir, "model.marker").write_text("save", encoding="utf-8")
+
+    def save_pretrained(self, output_dir):
+        Path(output_dir, "model.marker").write_text("save_pretrained", encoding="utf-8")
+
+
+def training_args(tmp_path):
+    return TrainingArguments(
+        output_dir=str(tmp_path / "output"),
+        report_to=[],
+    )
+
+
+def test_processing_class_and_legacy_tokenizer_are_accepted(tmp_path):
+    processor = RecordingProcessor("processing")
+    trainer = EncoderOnlyEmbedderTrainer(
+        model=SaveableModel(),
+        args=training_args(tmp_path),
+        processing_class=processor,
+    )
+    assert trainer.processing_class is processor
+
+    legacy_processor = RecordingProcessor("legacy")
+    legacy_trainer = EncoderOnlyEmbedderTrainer(
+        model=SaveableModel(),
+        args=training_args(tmp_path / "legacy"),
+        tokenizer=legacy_processor,
+    )
+    assert legacy_trainer.processing_class is legacy_processor
+
+
+def test_all_flagembedding_finetune_trainers_save_the_processing_class(tmp_path):
+    trainer_types = [
+        DecoderOnlyEmbedderTrainer,
+        DecoderOnlyEmbedderICLTrainer,
+        EncoderOnlyEmbedderTrainer,
+        EncoderOnlyEmbedderM3Trainer,
+        DecoderOnlyRerankerTrainer,
+        DecoderOnlyRerankerLayerwiseTrainer,
+        EncoderOnlyRerankerTrainer,
+    ]
+
+    for trainer_type in trainer_types:
+        output_dir = tmp_path / trainer_type.__name__
+        processor = RecordingProcessor(trainer_type.__name__)
+        trainer = trainer_type(
+            model=SaveableModel(),
+            args=training_args(output_dir),
+            processing_class=processor,
+        )
+
+        trainer._save(str(output_dir))
+
+        assert Path(output_dir, "processor.marker").read_text(encoding="utf-8") == trainer_type.__name__
+        assert Path(output_dir, "training_args.bin").exists()
+
+
+def test_reranker_runners_pass_processing_class(tmp_path):
+    runner_types = [
+        EncoderOnlyRerankerRunner,
+        DecoderOnlyRerankerRunner,
+        DecoderOnlyRerankerLayerwiseRunner,
+    ]
+
+    for runner_type in runner_types:
+        runner = runner_type.__new__(runner_type)
+        processor = RecordingProcessor(runner_type.__name__)
+        runner.model = SaveableModel()
+        runner.training_args = training_args(tmp_path / runner_type.__name__)
+        runner.train_dataset = None
+        runner.data_collator = None
+        runner.tokenizer = processor
+
+        trainer = runner.load_trainer()
+
+        assert trainer.processing_class is processor
+
+
+def test_legacy_transformers_trainer_constructor_path(monkeypatch, tmp_path):
+    """The compatibility shim uses tokenizer= with pre-v5 Trainer APIs."""
+    captured = {}
+
+    def legacy_init(self, *args, tokenizer=None, **kwargs):
+        captured["tokenizer"] = tokenizer
+        self.tokenizer = tokenizer
+        self.model = kwargs.get("model")
+        self.args = kwargs.get("args")
+
+    monkeypatch.setattr(Trainer, "__init__", legacy_init)
+
+    from FlagEmbedding.abc.finetune.embedder.AbsTrainer import AbsEmbedderTrainer
+
+    class LegacyTrainer(AbsEmbedderTrainer):
+        def _save(self, output_dir=None, state_dict=None):
+            pass
+
+    processor = RecordingProcessor("legacy-transformers")
+    LegacyTrainer(
+        model=SaveableModel(),
+        args=training_args(tmp_path),
+        processing_class=processor,
+    )
+
+    assert captured["tokenizer"] is processor