[grpo] remove data collator to top-level to avoid pickle error in spawn mode (#4582)

* remove data collator to top-level

* comment

Author: hjh0119

swift/trainers/rlhf_trainer/grpo_trainer.py

@@ -74,6 +74,10 @@ class DataCache:
     outputs: List[Dict] = field(default_factory=list)
 
 
+def identity_data_collator(features):
+    return features
+
+
 class GRPOTrainer(RLHFTrainerMixin, SwiftMixin, HFGRPOTrainer):
     executor = concurrent.futures.ThreadPoolExecutor(max_workers=1)
 
@@ -173,10 +177,7 @@ def __init__(self,
         self.completion_length_limit_scope = args.completion_length_limit_scope
         model.warnings_issued['estimate_tokens'] = True
 
-        def data_collator(features):  # No data collation is needed in GRPO
-            return features
-
-        kwargs['data_collator'] = data_collator
+        kwargs['data_collator'] = identity_data_collator  # No data collation is needed in GRPO
         self.shuffle_dataset = args.dataset_shuffle
 
         self.use_vllm = args.use_vllm