Doc

src/transformers/trainer_pt_utils.py

@@ -1288,6 +1288,17 @@ class AcceleratorConfig:
             "multiple different seeds to compare. Should also be ran with [`~utils.set_seed`] for the best results."
         },
     )
+
+    non_blocking: Optional[bool] = field(
+        default=False,
+        metadata={
+            "help": "Whether to use non-blocking CUDA calls to help minimize synchronization during "
+            "distributed training with prepared `DataLoader` inputs being moved to device. "
+            "Best if used with `pin_memory=True` in the `TrainingArguments`. Requires accelerate "
+            "v0.30.0."
+        },
+    )
+
     gradient_accumulation_kwargs: Optional[Dict] = field(
         default=None,
         metadata={
@@ -1302,15 +1313,6 @@ class AcceleratorConfig:
         },
     )
 
-    non_blocking: Optional[bool] = field(
-        default=False,
-        metadata={
-            "help": "Whether to use non-blocking CUDA calls to help minimize synchronization during "
-            "distributed training with prepared `DataLoader` inputs being moved to device. "
-            "Best if used with `pin_memory=True` in the `TrainingArguments`."
-        },
-    )
-
     @classmethod
     def from_json_file(cls, json_file):
         # Check if exists