diff --git a/vllm/v1/attention/backends/mla/rocm_aiter_mla_sparse.py b/vllm/v1/attention/backends/mla/rocm_aiter_mla_sparse.py index e1365f4d30..69b6e342cd 100644 --- a/vllm/v1/attention/backends/mla/rocm_aiter_mla_sparse.py +++ b/vllm/v1/attention/backends/mla/rocm_aiter_mla_sparse.py @@ -808,6 +808,11 @@ class ROCMAiterMLASparseImpl( (q_concat_shape, vllm_config.model_config.dtype), ) + def record_logical_topk_ready(self) -> None: + # This impl shares the top-k indices buffer via SharedTopkIndicesBuffer + # but does not participate in sparse-MLA index groups. + pass + def _forward_mla( self, layer: AttentionLayer,