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

[V0 deprecation] Remove _VLLM_V1 suffixes from attention backend names (#25489)

Browse Source 
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Signed-off-by: Matthew Bonanni <mbonanni001@gmail.com>
This commit is contained in:
Matthew Bonanni
2025-09-25 13:37:50 -04:00
committed by GitHub
parent 71b25b0d48
commit 3468f17ebe
42 changed files with 131 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/models/test_initialization.py
View File

@@ -84,7 +84,7 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
# FIXME: A hack to bypass FA3 assertion because our CI's L4 GPU
# has cc==8.9 which hasn't supported FA3 yet. Remove this hack when
# L4 supports FA3.
m.setenv("VLLM_ATTENTION_BACKEND", "TRITON_ATTN_VLLM_V1")
m.setenv("VLLM_ATTENTION_BACKEND", "TRITON_ATTN")
if model_arch == "WhisperForConditionalGeneration":
m.setenv("VLLM_WORKER_MULTIPROC_METHOD", "spawn")
LLM(