[CI] Make AttentionBackend interface compatible to fix broken CI (#1893)

vLLM commit
752c6ade2e
removed `blocksparse_params` for attention backend. This PR does the
same change to make CI happy.


- vLLM version: v0.9.2
- vLLM main:
9499e26e2a

---------

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
Co-authored-by: Yikun Jiang <yikunkero@gmail.com>

vllm_ascend/attention/attention_v1.py

@@ -31,7 +31,7 @@ from vllm.v1.worker.gpu_input_batch import InputBatch
 
 from vllm_ascend.ops.attention import vanilla_chunked_prefill
 from vllm_ascend.utils import (ACL_FORMAT_FRACTAL_NZ, aligned_16, is_310p,
-                               nd_to_nz_2d, nd_to_nz_spec)
+                               nd_to_nz_2d, nd_to_nz_spec, vllm_version_is)
 
 
 class AscendAttentionBackend(AttentionBackend):
@@ -43,6 +43,8 @@ class AscendAttentionBackend(AttentionBackend):
 
     @staticmethod
     def get_impl_cls() -> Type["AscendAttentionBackendImpl"]:
+        if vllm_version_is("0.9.2"):
+            return AscendAttentionBackendImpl092
         return AscendAttentionBackendImpl
 
     @staticmethod
@@ -222,7 +224,6 @@ class AscendAttentionBackendImpl(AttentionImpl):
         alibi_slopes: Optional[List[float]],
         sliding_window: Optional[int],
         kv_cache_dtype: str,
-        blocksparse_params: Optional[Dict[str, Any]] = None,
         logits_soft_cap: Optional[float] = None,
         attn_type: str = AttentionType.DECODER,
         kv_sharing_target_layer_name: Optional[str] = None,
@@ -437,6 +438,38 @@ class AscendAttentionBackendImpl(AttentionImpl):
         return output.view(num_tokens, self.hidden_size)
 
 
+class AscendAttentionBackendImpl092(AscendAttentionBackendImpl):
+
+    def __init__(
+        self,
+        num_heads: int,
+        head_size: int,
+        scale: float,
+        num_kv_heads: int,
+        alibi_slopes: Optional[List[float]],
+        sliding_window: Optional[int],
+        kv_cache_dtype: str,
+        blocksparse_params: Optional[Dict[str, Any]] = None,
+        logits_soft_cap: Optional[float] = None,
+        attn_type: str = AttentionType.DECODER,
+        kv_sharing_target_layer_name: Optional[str] = None,
+        use_irope: bool = False,
+    ) -> None:
+        super().__init__(
+            num_heads=num_heads,
+            head_size=head_size,
+            scale=scale,
+            num_kv_heads=num_kv_heads,
+            alibi_slopes=alibi_slopes,
+            sliding_window=sliding_window,
+            kv_cache_dtype=kv_cache_dtype,
+            logits_soft_cap=logits_soft_cap,
+            attn_type=attn_type,
+            kv_sharing_target_layer_name=kv_sharing_target_layer_name,
+            use_irope=use_irope,
+        )
+
+
 def unified_ascend_attention_with_output(
     query: torch.Tensor,
     key: torch.Tensor,