[Attention][V0 Deprecation] Deprecate accept output buffer (#39125)

Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>

vllm/v1/attention/backends/tree_attn.py

@@ -30,7 +30,6 @@ logger = init_logger(__name__)
 
 
 class TreeAttentionBackend(AttentionBackend):
-    accept_output_buffer: bool = True
     supported_dtypes: ClassVar[list[torch.dtype]] = [torch.float16, torch.bfloat16]
     supported_kv_cache_dtypes: ClassVar[list[CacheDType]] = [
         "auto",
@@ -368,7 +367,7 @@ class TreeAttentionImpl(AttentionImpl):
         value: torch.Tensor,
         kv_cache: torch.Tensor,
         attn_metadata: TreeAttentionMetadata,
-        output: torch.Tensor | None = None,
+        output: torch.Tensor,
         output_scale: torch.Tensor | None = None,
         output_block_scale: torch.Tensor | None = None,
     ) -> torch.Tensor:
@@ -384,8 +383,6 @@ class TreeAttentionImpl(AttentionImpl):
         Returns:
             shape = [num_tokens, num_heads * head_size]
         """
-        assert output is not None, "Output tensor must be provided."
-
         if output_scale is not None or output_block_scale is not None:
             raise NotImplementedError(
                 "fused output quantization is not yet supported for TreeAttentionImpl"

vllm/v1/attention/backends/triton_attn.py

@@ -262,7 +262,6 @@ class TritonAttentionMetadataBuilder(AttentionMetadataBuilder[TritonAttentionMet
 
 
 class TritonAttentionBackend(AttentionBackend):
-    accept_output_buffer: bool = True
     supported_dtypes: ClassVar[list[torch.dtype]] = [
         torch.float16,
         torch.bfloat16,
@@ -504,7 +503,7 @@ class TritonAttentionImpl(AttentionImpl):
         value: torch.Tensor,
         kv_cache: torch.Tensor,
         attn_metadata: TritonAttentionMetadata,
-        output: torch.Tensor | None = None,
+        output: torch.Tensor,
         output_scale: torch.Tensor | None = None,
         output_block_scale: torch.Tensor | None = None,
     ) -> torch.Tensor:
@@ -520,8 +519,6 @@ class TritonAttentionImpl(AttentionImpl):
         Returns:
             shape = [num_tokens, num_heads * head_size]
         """
-        assert output is not None, "Output tensor must be provided."
-
         if output_block_scale is not None:
             raise NotImplementedError(
                 "fused block_scale output quantization is not yet supported"