diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index 242d234d..6dd17e73 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -285,6 +285,8 @@ def patch_mistral_nemo_config(config):
 
 # =============================================
 # Fix new Xformers versions TypeError: Multiple dispatch failed for 'torch._ops.aten.to.dtype_layout'
+accelerate_old_send_to_device = None
+accelerate_new_send_to_device = None
 if Version(xformers_version) >= Version("0.0.27"):
     import accelerate.utils.operations
     if hasattr(accelerate.utils.operations, "send_to_device") and \
diff --git a/unsloth/models/llama.py b/unsloth/models/llama.py
index 3fcb8a76..39998127 100644
--- a/unsloth/models/llama.py
+++ b/unsloth/models/llama.py
@@ -1371,8 +1371,10 @@ def _fast_generate(*args, **kwargs):
         internal_model._flag_for_generation = True
 
         # Must patch accelerate for Xformers
-        import accelerate.utils.operations
-        accelerate.utils.operations.send_to_device = accelerate_new_send_to_device
+        if accelerate_new_send_to_device is not None:
+            import accelerate.utils.operations
+            accelerate.utils.operations.send_to_device = accelerate_new_send_to_device
+        pass
 
         # For newer HF
         kwargs["cache_implementation"] = "dynamic"
@@ -1411,7 +1413,9 @@ def _fast_generate(*args, **kwargs):
         if hasattr(internal_model, "_flag_for_generation"): del internal_model._flag_for_generation
 
         # Return accelerate back
-        accelerate.utils.operations.send_to_device = accelerate_old_send_to_device
+        if accelerate_new_send_to_device is not None:
+            accelerate.utils.operations.send_to_device = accelerate_old_send_to_device
+        pass
 
         return output
     pass