diff --git a/colossalai/shardformer/layer/_operation.py b/colossalai/shardformer/layer/_operation.py
index 0d8c3d453ce1..8fd92a2ed6b8 100644
--- a/colossalai/shardformer/layer/_operation.py
+++ b/colossalai/shardformer/layer/_operation.py
@@ -473,16 +473,17 @@ def forward(ctx, input_, dim, process_group):
     @staticmethod
     def backward(ctx, grad_output):
         return _split(grad_output, ctx.dim, ctx.process_group), None, None
-    
+
 
 class HookParameter(torch.autograd.Function):
     """In order to be hooked into Gemini's '__torch_function__', adding a view operation to weight and bias. Used in FusedLayerNorm"""
+
     @staticmethod
     def forward(ctx, input, weight, bias):
         ctx.save_for_backward(weight, bias)
         output = input
         return output
-    
+
     @staticmethod
     def backward(ctx, grad_output):
         weight, bias = ctx.saved_tensors
@@ -491,13 +492,12 @@ def backward(ctx, grad_output):
         if bias is not None:
             bias = bias.view(bias.shape)
         return grad_output, None, None
-    
+
 
 def hook_paramter_in_backward(input, weight=None, bias=None):
     return HookParameter.apply(input, weight, bias)
 
 
-
 def _reduce(input_, process_group):
     # skip if only one rank involved
     if dist.get_world_size(process_group) == 1:
@@ -522,7 +522,7 @@ def _split(input_, dim=-1, process_group=None):
 
     tensor_list = torch.split(input_, dim_size // world_size, dim=dim)
     rank = dist.get_rank(process_group)
-    output = tensor_list[rank].contiguous()
+    output = tensor_list[rank].clone().contiguous()
 
     return output
 
diff --git a/colossalai/tensor/d_tensor/comm_spec.py b/colossalai/tensor/d_tensor/comm_spec.py
index 8f5b52aab8f8..fc017c663a81 100644
--- a/colossalai/tensor/d_tensor/comm_spec.py
+++ b/colossalai/tensor/d_tensor/comm_spec.py
@@ -112,7 +112,7 @@ def _split(tensor: torch.Tensor, comm_spec: CommSpec):
     dim = comm_spec.shard_dim
     length = tensor.shape[comm_spec.shard_dim] // dist.get_world_size(process_group)
     start = length * dist.get_rank(process_group)
-    output = torch.narrow(tensor, dim, start, length).contiguous()
+    output = torch.narrow(tensor, dim, start, length).clone().contiguous()
     return output