diff --git a/vllm_ascend/torchair/quantization/torchair_w8a8_dynamic.py b/vllm_ascend/torchair/quantization/torchair_w8a8_dynamic.py
index ceba1c4..5bd622e 100644
--- a/vllm_ascend/torchair/quantization/torchair_w8a8_dynamic.py
+++ b/vllm_ascend/torchair/quantization/torchair_w8a8_dynamic.py
@@ -459,7 +459,7 @@ def torchair_fused_experts_with_all2all(
         token_counts_combined = token_counts_combined.view(
             2, ep_group.world_size, -1).sum(dim=2)
         token_counts_combined_cpu = token_counts_combined.to(
-            torch.device("cpu"), non_blocking=True).numpy()
+            torch.device("cpu"), non_blocking=False).numpy()
         all_tokens = gather_sizes.sum()
 
         gathered_tokens = quantized_tokens.new_empty(all_tokens.item(),