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

Quick Fix by adding conditional import for flash_attn_varlen_func in flash_attn (#20143)

Browse Source 
Signed-off-by: Chendi.Xue <chendi.xue@intel.com>
This commit is contained in:
Chendi.Xue
2025-06-27 00:48:13 -05:00
committed by GitHub
parent 6e244ae091
commit dec197e3e5
2 changed files with 11 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/v1/attention/backends/flash_attn.py
View File

@@ -14,10 +14,14 @@ from vllm.attention.backends.abstract import (AttentionBackend, AttentionImpl,
from vllm.attention.layer import Attention
from vllm.attention.ops.merge_attn_states import merge_attn_states
from vllm.attention.utils.fa_utils import (flash_attn_supports_fp8,
flash_attn_varlen_func,
get_flash_attn_version,
get_scheduler_metadata,
reshape_and_cache_flash)
is_flash_attn_varlen_func_available)
if is_flash_attn_varlen_func_available():
from vllm.attention.utils.fa_utils import (flash_attn_varlen_func,
get_scheduler_metadata,
reshape_and_cache_flash)
from vllm.config import VllmConfig, get_layers_from_vllm_config
from vllm.logger import init_logger
from vllm.utils import cdiv