diff --git a/vllm/model_executor/models/qwen3_5.py b/vllm/model_executor/models/qwen3_5.py
index e5967c122..78dda9ff4 100644
--- a/vllm/model_executor/models/qwen3_5.py
+++ b/vllm/model_executor/models/qwen3_5.py
@@ -182,8 +182,8 @@ class Qwen3_5GatedDeltaNet(Qwen3NextGatedDeltaNet):
         # ============================================================
         mixed_qkvz, ba = torch.ops.vllm.gdn_in_proj(
             hidden_states,
-            self.in_proj_qkvz.weight.shape[0],
-            self.in_proj_ba.weight.shape[0],
+            sum(self.in_proj_qkvz.output_sizes) // self.tp_size,
+            sum(self.in_proj_ba.output_sizes) // self.tp_size,
             self.prefix,
         )
         qkv_size = (self.key_dim * 2 + self.value_dim) // self.tp_size
diff --git a/vllm/model_executor/models/qwen3_next.py b/vllm/model_executor/models/qwen3_next.py
index 7aaded7ae..bf59c0c11 100644
--- a/vllm/model_executor/models/qwen3_next.py
+++ b/vllm/model_executor/models/qwen3_next.py
@@ -660,8 +660,8 @@ class Qwen3NextGatedDeltaNet(nn.Module, MambaBase):
         # ============================================================
         projected_states_qkvz, projected_states_ba = torch.ops.vllm.gdn_in_proj(
             hidden_states,
-            self.in_proj_qkvz.weight.shape[0],
-            self.in_proj_ba.weight.shape[0],
+            sum(self.in_proj_qkvz.output_sizes) // self.tp_size,
+            sum(self.in_proj_ba.output_sizes) // self.tp_size,
             self.prefix,
         )
         query, key, value, z, b, a = self.fix_query_key_value_ordering(