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

VLLM_USE_TRITON_FLASH_ATTN V0 variable deprecation (#27611)

Browse Source 
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Signed-off-by: Andreas Karatzas <Andreas.Karatzas@amd.com>
This commit is contained in:
Andreas Karatzas
2025-11-11 20:34:36 -06:00
committed by GitHub
parent 7f829be7d3
commit 9f0247cfa4
15 changed files with 12 additions and 1588 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/models/language/pooling/test_embedding.py
View File

@@ -4,7 +4,6 @@
import pytest
from vllm.config import PoolerConfig
from vllm.platforms import current_platform
from ...utils import check_embeddings_close
@@ -51,13 +50,7 @@ def test_models(
vllm_runner,
example_prompts,
model,
monkeypatch,
) -> None:
if model == "BAAI/bge-multilingual-gemma2" and current_platform.is_rocm():
# ROCm Triton FA does not currently support sliding window attention
# switch to use ROCm CK FA backend
monkeypatch.setenv("VLLM_USE_TRITON_FLASH_ATTN", "False")
vllm_extra_kwargs = {}
if model == "ssmits/Qwen2-7B-Instruct-embed-base":
vllm_extra_kwargs["pooler_config"] = PoolerConfig(