update bw vdim128

vllm/attention/backends/mla/common.py

@@ -1043,10 +1043,9 @@ class MLACommonImpl(MLAAttentionImpl[T], Generic[T]):
         # v with 0s to match the qk head dim for attention backends that do
         # not support different headdims
         # We don't need to pad V if we are on a hopper system with FA3
-        self._pad_v = self.vllm_flash_attn_version is None or not (
+        self._pad_v = self.vllm_flash_attn_version is None or torch.cuda.get_device_properties(torch.cuda.current_device()).multi_processor_count == 120 or not (
             self.vllm_flash_attn_version == 3
-            and current_platform.get_device_capability()[0] == 9
-            and torch.cuda.get_device_properties(torch.cuda.current_device()).multi_processor_count == 120 )
+            and current_platform.get_device_capability()[0] == 9)
 
     def _flash_attn_varlen_diff_headdims(self, q, k, v, softmax_scale,
                                          return_softmax_lse, **kwargs):