[Trainer] Fix default data collator (#30142)

* Fix data collator

* Support feature extractors as well

src/transformers/trainer.py

@@ -58,6 +58,7 @@
 from .configuration_utils import PretrainedConfig
 from .data.data_collator import DataCollator, DataCollatorWithPadding, default_data_collator
 from .debug_utils import DebugOption, DebugUnderflowOverflow
+from .feature_extraction_sequence_utils import SequenceFeatureExtractor
 from .hyperparameter_search import ALL_HYPERPARAMETER_SEARCH_BACKENDS, default_hp_search_backend
 from .image_processing_utils import BaseImageProcessor
 from .integrations.deepspeed import deepspeed_init, deepspeed_load_checkpoint, is_deepspeed_available
@@ -492,7 +493,11 @@ def __init__(
         ):
             self.place_model_on_device = False
 
-        default_collator = DataCollatorWithPadding(tokenizer) if tokenizer is not None else default_data_collator
+        default_collator = (
+            DataCollatorWithPadding(tokenizer)
+            if tokenizer is not None and isinstance(tokenizer, (PreTrainedTokenizerBase, SequenceFeatureExtractor))
+            else default_data_collator
+        )
         self.data_collator = data_collator if data_collator is not None else default_collator
         self.train_dataset = train_dataset
         self.eval_dataset = eval_dataset