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

[Bug][DSV3.2] Always prepare metadata for DeepGEMM Sparse Attention (#35075)

Browse Source 
Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>
This commit is contained in:
Benjamin Chislett
2026-02-24 17:55:22 -05:00
committed by GitHub
parent ea97750414
commit 9fa5b25a23
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/v1/attention/backends/mla/indexer.py
View File

@@ -8,7 +8,7 @@ import torch
from vllm.config import VllmConfig
from vllm.logger import init_logger
from vllm.platforms import current_platform
from vllm.utils.deep_gemm import get_paged_mqa_logits_metadata, is_deep_gemm_supported
from vllm.utils.deep_gemm import get_paged_mqa_logits_metadata, has_deep_gemm
from vllm.v1.attention.backend import (
AttentionBackend,
AttentionCGSupport,
@@ -342,7 +342,9 @@ class DeepseekV32IndexerMetadataBuilder(AttentionMetadataBuilder):
offsets = None
seq_lens = common_attn_metadata.seq_lens[:num_decodes]
if is_deep_gemm_supported():
# DeepGEMM is required for the paged MQA logits on CUDA devices
if current_platform.is_cuda() and has_deep_gemm():
self.scheduler_metadata_buffer[:] = get_paged_mqa_logits_metadata(
seq_lens, self.kv_cache_spec.block_size, self.num_sms
)