[Attention] Unify mamba and attention backend selection (#23171)

Signed-off-by: Ayush Satyam <ayushsatyam146@gmail.com>

vllm/model_executor/layers/mamba/short_conv.py

@@ -1,7 +1,10 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
-from typing import Optional
+from typing import TYPE_CHECKING, Optional
+
+if TYPE_CHECKING:
+    from vllm.attention.backends.abstract import AttentionBackend
 
 import torch
 
@@ -232,6 +235,11 @@ class ShortConv(MambaBase, CustomOp):
     def mamba_type(self) -> str:
         return "short_conv"
 
+    def get_attn_backend(self) -> type["AttentionBackend"]:
+        from vllm.v1.attention.backends.short_conv_attn import (
+            ShortConvAttentionBackend)
+        return ShortConvAttentionBackend
+
 
 def short_conv(
     hidden_states: torch.Tensor,