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

[Attention] Refactor CUDA attention backend selection logic (#24794)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
This commit is contained in:
Matthew Bonanni
2025-11-11 06:40:44 -06:00
committed by GitHub
parent 2e78150d24
commit b30dfa03c5
61 changed files with 1338 additions and 1002 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tests/config/test_multimodal_config.py
View File

@@ -3,13 +3,13 @@
import pytest
from vllm.attention.backends.registry import _Backend
from vllm.attention.backends.registry import AttentionBackendEnum
from vllm.config.multimodal import MultiModalConfig
def test_mm_encoder_attn_backend_str_conversion():
config = MultiModalConfig(mm_encoder_attn_backend="FLASH_ATTN")
assert config.mm_encoder_attn_backend == _Backend.FLASH_ATTN
assert config.mm_encoder_attn_backend == AttentionBackendEnum.FLASH_ATTN
def test_mm_encoder_attn_backend_invalid():
@@ -20,6 +20,6 @@ def test_mm_encoder_attn_backend_invalid():
def test_mm_encoder_attn_backend_hash_updates():
base_hash = MultiModalConfig().compute_hash()
overridden_hash = MultiModalConfig(
mm_encoder_attn_backend=_Backend.FLASH_ATTN
mm_encoder_attn_backend=AttentionBackendEnum.FLASH_ATTN
).compute_hash()
assert base_hash != overridden_hash