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

Remove V0 Encoder-Decoder Support (#24907)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk@thinkingmachines.ai>
This commit is contained in:
Woosuk Kwon
2025-09-15 21:17:14 -07:00
committed by GitHub
parent 5206ab20ba
commit 759ef49b15
47 changed files with 13 additions and 9661 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/models/test_initialization.py
View File

@@ -92,10 +92,6 @@ def can_initialize(model_arch: str, monkeypatch: pytest.MonkeyPatch,
# 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")
if model_arch == "Florence2ForConditionalGeneration":
# An encoder-decoder model that's V0-only. Just skip it
# since V0 is about to be removed.
pytest.skip("Skipping Florence2ForConditionalGeneration")
if model_arch == "WhisperForConditionalGeneration":
m.setenv("VLLM_WORKER_MULTIPROC_METHOD", "spawn")
LLM(