🎯 Don't use getattr to get gradient_checkpointing

trl/trainer/bco_trainer.py

@@ -431,7 +431,7 @@ def __init__(
                     prepare_model_kwargs["gradient_checkpointing_kwargs"] = args.gradient_checkpointing_kwargs
 
                 model = prepare_model_for_kbit_training(model, **prepare_model_kwargs)
-            elif getattr(args, "gradient_checkpointing", False):
+            elif args.gradient_checkpointing:
                 # For backward compatibility with older versions of transformers
                 if hasattr(model, "enable_input_require_grads"):
                     model.enable_input_require_grads()
@@ -452,7 +452,7 @@ def make_inputs_require_grad(module, input, output):
         # For models that use gradient_checkpointing, we need to attach a hook that enables input
         # to explicitly have `requires_grad=True`, otherwise training will either silently
         # fail or completely fail.
-        elif getattr(args, "gradient_checkpointing", False):
+        elif args.gradient_checkpointing:
             # For backward compatibility with older versions of transformers
             if hasattr(model, "enable_input_require_grads"):
                 model.enable_input_require_grads()

trl/trainer/cpo_trainer.py

@@ -172,7 +172,7 @@ def __init__(
                     prepare_model_kwargs["gradient_checkpointing_kwargs"] = args.gradient_checkpointing_kwargs
 
                 model = prepare_model_for_kbit_training(model, **prepare_model_kwargs)
-            elif getattr(args, "gradient_checkpointing", False):
+            elif args.gradient_checkpointing:
                 # For backward compatibility with older versions of transformers
                 if hasattr(model, "enable_input_require_grads"):
                     model.enable_input_require_grads()
@@ -193,7 +193,7 @@ def make_inputs_require_grad(module, input, output):
         # For models that use gradient_checkpointing, we need to attach a hook that enables input
         # to explicitly have `requires_grad=True`, otherwise training will either silently
         # fail or completely fail.
-        elif getattr(args, "gradient_checkpointing", False):
+        elif args.gradient_checkpointing:
             # For backward compatibility with older versions of transformers
             if hasattr(model, "enable_input_require_grads"):
                 model.enable_input_require_grads()

trl/trainer/dpo_trainer.py

@@ -302,7 +302,7 @@ def __init__(
                     prepare_model_kwargs["gradient_checkpointing_kwargs"] = args.gradient_checkpointing_kwargs
 
                 model = prepare_model_for_kbit_training(model, **prepare_model_kwargs)
-            elif getattr(args, "gradient_checkpointing", False):
+            elif args.gradient_checkpointing:
                 # For backward compatibility with older versions of transformers
                 if hasattr(model, "enable_input_require_grads"):
                     model.enable_input_require_grads()
@@ -323,7 +323,7 @@ def make_inputs_require_grad(module, input, output):
         # For models that use gradient_checkpointing, we need to attach a hook that enables input
         # to explicitly have `requires_grad=True`, otherwise training will either silently
         # fail or completely fail.
-        elif getattr(args, "gradient_checkpointing", False):
+        elif args.gradient_checkpointing:
             # For backward compatibility with older versions of transformers
             if hasattr(model, "enable_input_require_grads"):
                 model.enable_input_require_grads()

trl/trainer/kto_trainer.py

@@ -416,7 +416,7 @@ def __init__(
                     prepare_model_kwargs["gradient_checkpointing_kwargs"] = args.gradient_checkpointing_kwargs
 
                 model = prepare_model_for_kbit_training(model, **prepare_model_kwargs)
-            elif getattr(args, "gradient_checkpointing", False):
+            elif args.gradient_checkpointing:
                 # For backward compatibility with older versions of transformers
                 if hasattr(model, "enable_input_require_grads"):
                     model.enable_input_require_grads()
@@ -437,7 +437,7 @@ def make_inputs_require_grad(module, input, output):
         # For models that use gradient_checkpointing, we need to attach a hook that enables input
         # to explicitly have `requires_grad=True`, otherwise training will either silently
         # fail or completely fail.
-        elif getattr(args, "gradient_checkpointing", False):
+        elif args.gradient_checkpointing:
             # For backward compatibility with older versions of transformers
             if hasattr(model, "enable_input_require_grads"):
                 model.enable_input_require_grads()

trl/trainer/orpo_trainer.py

@@ -176,7 +176,7 @@ def __init__(
                     prepare_model_kwargs["gradient_checkpointing_kwargs"] = args.gradient_checkpointing_kwargs
 
                 model = prepare_model_for_kbit_training(model, **prepare_model_kwargs)
-            elif getattr(args, "gradient_checkpointing", False):
+            elif args.gradient_checkpointing:
                 # For backward compatibility with older versions of transformers
                 if hasattr(model, "enable_input_require_grads"):
                     model.enable_input_require_grads()
@@ -197,7 +197,7 @@ def make_inputs_require_grad(module, input, output):
         # For models that use gradient_checkpointing, we need to attach a hook that enables input
         # to explicitly have `requires_grad=True`, otherwise training will either silently
         # fail or completely fail.
-        elif getattr(args, "gradient_checkpointing", False):
+        elif args.gradient_checkpointing:
             # For backward compatibility with older versions of transformers
             if hasattr(model, "enable_input_require_grads"):
                 model.enable_input_require_grads()