diff --git a/.buildkite/test-amd.yaml b/.buildkite/test-amd.yaml
index 9561fec5e0..cca01e45ba 100644
--- a/.buildkite/test-amd.yaml
+++ b/.buildkite/test-amd.yaml
@@ -4022,7 +4022,7 @@ steps:
 #------------------------------------------------------  mi355 · weight_loading  -------------------------------------------------------#
 
 - label: ":amd: (MI355) Weight Loading Multi-GPU" # TBD
-  timeout_in_minutes: 30
+  timeout_in_minutes: 40
   mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
   dind: false
   agent_pool: mi355_2
diff --git a/vllm/model_executor/layers/linear.py b/vllm/model_executor/layers/linear.py
index cd1c0e69d2..2e77fac06c 100644
--- a/vllm/model_executor/layers/linear.py
+++ b/vllm/model_executor/layers/linear.py
@@ -380,7 +380,8 @@ class ReplicatedLinear(LinearBase):
 
         if bias:
             self.bias = Parameter(
-                torch.empty(self.output_size, dtype=self.params_dtype)
+                torch.empty(self.output_size, dtype=self.params_dtype),
+                requires_grad=False,
             )
             set_weight_attrs(
                 self.bias,
@@ -525,7 +526,8 @@ class ColumnParallelLinear(LinearBase):
 
         if bias:
             self.bias = Parameter(
-                torch.empty(self.output_size_per_partition, dtype=params_dtype)
+                torch.empty(self.output_size_per_partition, dtype=params_dtype),
+                requires_grad=False,
             )
             set_weight_attrs(
                 self.bias,
@@ -1708,7 +1710,10 @@ class RowParallelLinear(LinearBase):
             )
 
         if bias:
-            self.bias = Parameter(torch.empty(self.output_size, dtype=params_dtype))
+            self.bias = Parameter(
+                torch.empty(self.output_size, dtype=params_dtype),
+                requires_grad=False,
+            )
             set_weight_attrs(
                 self.bias,
                 {