diff --git a/vllm_ascend/attention/context_parallel/mla_cp.py b/vllm_ascend/attention/context_parallel/mla_cp.py
index 539f2a330e6..87413f5b692 100644
--- a/vllm_ascend/attention/context_parallel/mla_cp.py
+++ b/vllm_ascend/attention/context_parallel/mla_cp.py
@@ -74,7 +74,7 @@ def build(
         fast_build: bool = False,
     ) -> AscendMLAMetadata:
         metadata_cls = super().build(common_prefix_len, common_attn_metadata)
-        if self.num_prefills == 0 and self.pcp_size > 1:
+        if self.pcp_size > 1:
             self.slot_mapping[:self.
                               num_decode_tokens] = self.slot_mapping[:self.
                                                                      num_decode_tokens