xx

Author: wangxiaoxin-sherie

vllm_ascend/worker/model_runner_v1.py

@@ -2710,9 +2710,9 @@ def _build_dummy_attn_metadata(
 
             assigned_mask_dim = 2048
             self.attn_mask = torch.triu(torch.ones(assigned_mask_dim,
-                                                       assigned_mask_dim),
-                                            diagonal=1).to(torch.int8).to(
-                                                self.device)
+                                                   assigned_mask_dim),
+                                        diagonal=1).to(torch.int8).to(
+                                            self.device)
 
             num_computed_tokens_cpu = (
                 self.input_batch.num_computed_tokens_cpu_tensor[:num_reqs])