diff --git a/colossalai/shardformer/layer/linear.py b/colossalai/shardformer/layer/linear.py
index 085609c01..d77dd4965 100644
--- a/colossalai/shardformer/layer/linear.py
+++ b/colossalai/shardformer/layer/linear.py
@@ -202,21 +202,21 @@ class Linear1D_Col(ParallelModule):
         # Matrix multiply.
         bias = self.bias if not self.skip_bias_add else None
 
-        if self.seq_parallel_mode is None:
-            output_parallel = linear_with_async_comm(
-                input_parallel, self.weight, bias, self.process_group, True, fp8_communication=self.fp8_communication
-            )
-        elif self.seq_parallel_mode == "split_gather":
+        if self.seq_parallel_mode == "split_gather":
             input_parallel = gather_forward_reducescatter_backward(
                 input_parallel, self.process_group, self.seq_parallel_dim, fp8_communication=self.fp8_communication
             )
-            output_parallel = linear_with_async_comm(input_parallel, self.weight, bias, self.process_group, False)
+            output_parallel = linear_with_async_comm(
+                input_parallel, self.weight, bias, self.process_group, False, fp8_communication=self.fp8_communication
+            )
         elif self.seq_parallel_mode == "ring":
             output_parallel = linear_gather_forward_reducescatter_backward(
                 input_parallel, self.weight, bias, self.process_group, True, self.seq_parallel_dim, self.overlap, True
             )
         else:
-            output_parallel = linear_with_async_comm(input_parallel, self.weight, bias, self.process_group, True)
+            output_parallel = linear_with_async_comm(
+                input_parallel, self.weight, bias, self.process_group, True, fp8_communication=self.fp8_communication
+            )
 
         if self.gather_output:
             # All-gather across the partitions.