2

vllm/attention/backends/openvino.py

@@ -1,5 +1,5 @@
 from dataclasses import dataclass
-from typing import List, Optional, Tuple
+from typing import List, Tuple
 
 import openvino as ov
 import torch
@@ -21,7 +21,11 @@ def get_impl_cls():
         raise NotImplementedError
 
     @staticmethod
-    def make_metadata(*args, **kwargs) -> "OpenVINOAttentionMetadata":
+    def make_metadata(*args, **kwargs) -> "AttentionMetadata":
+        raise NotImplementedError
+
+    @staticmethod
+    def make_openvino_metadata(*args, **kwargs) -> "OpenVINOAttentionMetadata":
         return OpenVINOAttentionMetadata(*args, **kwargs)
 
     @staticmethod
@@ -55,21 +59,11 @@ def copy_blocks(
 
 
 @dataclass
-class OpenVINOAttentionMetadata(AttentionMetadata):
+class OpenVINOAttentionMetadata:
     """Metadata for OpenVINOAttentionBackend.
     """
     past_lens: torch.Tensor
     subsequence_begins: torch.Tensor
     block_indices: torch.Tensor
     block_indices_begins: torch.Tensor
     max_context_len: torch.Tensor
-
-    @property
-    def prefill_metadata(self) -> Optional["AttentionMetadata"]:
-        # OpenVINO uses its own metadata format
-        raise NotImplementedError
-
-    @property
-    def decode_metadata(self) -> Optional["AttentionMetadata"]:
-        # OpenVINO uses its own metadata format
-        raise NotImplementedError

vllm/worker/openvino_model_runner.py

@@ -254,7 +254,7 @@ def _prepare_model_input(
             max_context_len, dtype=torch.int32,
             device=self.device)  # type: ignore
 
-        attn_metadata = self.attn_backend.make_metadata(
+        attn_metadata = self.attn_backend.make_openvino_metadata(
             past_lens=past_lens_tensor,
             subsequence_begins=subsequence_begins_tensor,
             block_indices=block_indices_tensor,