[Revert] Remove DeepGEMM availability check in DeepseekV32IndexerMetadataBuilder (#38076)
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
This commit is contained in:
@@ -10,7 +10,6 @@ from vllm.platforms import current_platform
|
|||||||
from vllm.utils.deep_gemm import (
|
from vllm.utils.deep_gemm import (
|
||||||
get_paged_mqa_logits_metadata,
|
get_paged_mqa_logits_metadata,
|
||||||
has_deep_gemm,
|
has_deep_gemm,
|
||||||
is_deep_gemm_supported,
|
|
||||||
)
|
)
|
||||||
from vllm.utils.math_utils import cdiv
|
from vllm.utils.math_utils import cdiv
|
||||||
from vllm.utils.platform_utils import num_compute_units
|
from vllm.utils.platform_utils import num_compute_units
|
||||||
@@ -216,12 +215,6 @@ class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder):
|
|||||||
vllm_config: VllmConfig,
|
vllm_config: VllmConfig,
|
||||||
kv_cache_spec: AttentionSpec,
|
kv_cache_spec: AttentionSpec,
|
||||||
) -> AttentionCGSupport:
|
) -> AttentionCGSupport:
|
||||||
if not is_deep_gemm_supported():
|
|
||||||
logger.warning_once(
|
|
||||||
"DeepGEMM is not available. Disabling CUDA graph support "
|
|
||||||
"for sparse attention indexer. This may reduce performance.",
|
|
||||||
)
|
|
||||||
return AttentionCGSupport.NEVER
|
|
||||||
return AttentionCGSupport.UNIFORM_BATCH
|
return AttentionCGSupport.UNIFORM_BATCH
|
||||||
|
|
||||||
def __init__(self, *args, **kwargs):
|
def __init__(self, *args, **kwargs):
|
||||||
|
|||||||
Reference in New Issue
Block a user