[Bugfix][ROCm] Preserve AITER unified-attention metadata during graph replay (#53821)

Signed-off-by: andyluo7 <andy.luo@amd.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
andyluo7
2026-08-31 21:53:12 +08:00
committed by GitHub
co-authored by OpenAI Codex
parent c5d840ff6a
commit 76ff0cdff2
2 changed files with 59 additions and 3 deletions
@@ -15,7 +15,12 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
kFp8StaticTensorSym,
)
from vllm.utils.torch_utils import is_quantized_kv_cache
from vllm.v1.attention.backend import AttentionLayer, AttentionType, MultipleOf
from vllm.v1.attention.backend import (
AttentionLayer,
AttentionType,
CommonAttentionMetadata,
MultipleOf,
)
from vllm.v1.attention.backends.rocm_attn import (
RocmAttentionBackend,
RocmAttentionImpl,
@@ -27,6 +32,17 @@ from vllm.v1.kv_cache_interface import AttentionSpec, KVCacheLayout
logger = init_logger(__name__)
class RocmAiterUnifiedAttentionMetadataBuilder(RocmAttentionMetadataBuilder):
def build_for_cudagraph_capture(
self, common_attn_metadata: CommonAttentionMetadata
) -> RocmAttentionMetadata:
attn_metadata = self.build(0, common_attn_metadata)
# Avoid capturing the real maximum sequence length while preserving
# query_start_loc, which unified attention consumes during replay.
attn_metadata.seq_lens.fill_(1)
return attn_metadata
class RocmAiterUnifiedAttentionBackend(RocmAttentionBackend):
supported_dtypes: ClassVar[list[torch.dtype]] = [torch.float16, torch.bfloat16]
supported_kv_cache_dtypes: ClassVar[list[CacheDType]] = [
@@ -101,8 +117,8 @@ class RocmAiterUnifiedAttentionBackend(RocmAttentionBackend):
return False
@staticmethod
def get_builder_cls() -> type["RocmAttentionMetadataBuilder"]:
return RocmAttentionMetadataBuilder
def get_builder_cls() -> type["RocmAiterUnifiedAttentionMetadataBuilder"]:
return RocmAiterUnifiedAttentionMetadataBuilder
@classmethod
def supports_attn_type(cls, attn_type: str) -> bool: