biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Revert] Remove DeepGEMM availability check in DeepseekV32IndexerMetadataBuilder (#38076)

Browse Source 
Signed-off-by: chaunceyjiang <chaunceyjiang@gmail.com>
This commit is contained in:
Chauncey
2026-03-26 09:43:51 +08:00
committed by GitHub
parent 36f6aede23
commit 87f05d6880
1 changed files with 0 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/v1/attention/backends/mla/indexer.py
View File

@@ -10,7 +10,6 @@ from vllm.platforms import current_platform
from vllm.utils.deep_gemm import (
get_paged_mqa_logits_metadata,
has_deep_gemm,
is_deep_gemm_supported,
)
from vllm.utils.math_utils import cdiv
from vllm.utils.platform_utils import num_compute_units
@@ -216,12 +215,6 @@ class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder):
vllm_config: VllmConfig,
kv_cache_spec: AttentionSpec,
) -> 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
def __init__(self, *args, **kwargs):