diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
index e1871eaf8c3583..e585520cf4ac8e 100755
--- a/src/transformers/trainer.py
+++ b/src/transformers/trainer.py
@@ -2175,7 +2175,7 @@ def _inner_training_loop(
         grad_norm: Optional[float] = None
         self.control = self.callback_handler.on_train_begin(args, self.state, self.control)
 
-        if args.sanity_evaluation:
+        if args.eval_on_start:
             self._evaluate(trial, ignore_keys_for_eval, skip_scheduler=True)
 
         total_batched_samples = 0
diff --git a/src/transformers/training_args.py b/src/transformers/training_args.py
index b506e4b1507320..0b1f5bc9d1313c 100644
--- a/src/transformers/training_args.py
+++ b/src/transformers/training_args.py
@@ -772,8 +772,8 @@ class TrainingArguments:
             that takes a boolean argument `compute_result`, which when passed `True`, will trigger the final global
             summary statistics from the batch-level summary statistics you've accumulated over the evaluation set.
 
-        sanity_evaluation(`bool`, *optional*, defaults to `False`):
-            Whether or not to perform a sanity check to ensure that the validation steps works correctly. It will be performed before the training.
+        eval_on_start(`bool`, *optional*, defaults to `False`):
+            Whether to perform a evaluation step (sanity check) before the training to ensure the validation steps works correctly.
     """
 
     framework = "pt"
@@ -1457,7 +1457,7 @@ class TrainingArguments:
         metadata={"help": "Break eval metrics calculation into batches to save memory."},
     )
 
-    sanity_evaluation: bool = field(
+    eval_on_start: bool = field(
         default=False,
         metadata={
             "help": "Whether to run through the entire `evaluation` step at the very beginning of training as a sanity check."