fix ci

Author: JC-ut0

vllm_ascend/attention/attention_v1.py

@@ -218,9 +218,6 @@ def build_dummy_metadata(self, num_actual_tokens, num_reqs,
                              num_scheduled_tokens, attn_state):
         if attn_state == AscendAttentionState.DecodeOnly:
             # NOTE: We only need to pay attention to seq_lens_list and block_table here
-            common_attn_metadata = AscendCommonAttentionMetadata(
-                seq_lens=torch.empty_like(self.runner.seq_lens_cpu).fill_(2))
-
             block_table = self.runner.input_batch.block_table[0].block_table
             block_table[:num_reqs, 0] = torch.arange(1,
                                                      num_reqs + 1,
@@ -231,8 +228,8 @@ def build_dummy_metadata(self, num_actual_tokens, num_reqs,
             block_table[:num_reqs, :self.runner.max_num_blocks_per_req] = (
                 block_table[:num_reqs])
 
-            query_start_loc = common_attn_metadata.query_start_loc
-            seq_lens = common_attn_metadata.seq_lens
+            query_start_loc = None
+            seq_lens = torch.empty_like(self.runner.seq_lens_cpu).fill_(2)
             query_lens = self.runner.query_lens
             seq_lens_list = None