remove num_speculative_tokens init val

vllm/v1/attention/backends/utils.py

@@ -55,7 +55,7 @@ class CommonAttentionMetadata:
     """Longest query in batch"""
     num_rejected_tokens: list[int]
     """(batch_size,), record the rejected tokens number in cpu and gpu"""
-    num_speculative_tokens: int = 0
+    num_speculative_tokens: int
     """Number of speculative tokens"""
 
     block_table_tensor: torch.Tensor