diff --git a/vllm/v1/attention/backends/cpu_attn.py b/vllm/v1/attention/backends/cpu_attn.py index e28e200451..801a16d319 100644 --- a/vllm/v1/attention/backends/cpu_attn.py +++ b/vllm/v1/attention/backends/cpu_attn.py @@ -25,6 +25,7 @@ from vllm.v1.attention.backend import ( MultipleOf, ) from vllm.v1.attention.backends.utils import ( + KVCacheLayoutType, split_decodes_and_prefills, ) from vllm.v1.kv_cache_interface import AttentionSpec, CrossAttentionSpec @@ -94,6 +95,10 @@ class CPUAttentionBackend(AttentionBackend): ) -> tuple[int, ...]: return 2, num_blocks, num_kv_heads, block_size, head_size + @classmethod + def get_required_kv_cache_layout(cls) -> "KVCacheLayoutType | None": + return "HND" + @staticmethod def use_cascade_attention(*args, **kwargs) -> bool: return False