Remove V0 Encoder-Decoder Support (#24907)

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
-21
View File
@@ -9,24 +9,9 @@ from vllm import LLM
from vllm.engine.arg_utils import AsyncEngineArgs
from vllm.engine.async_llm_engine import AsyncLLMEngine
UNSUPPORTED_MODELS_V1 = [
"facebook/bart-large-cnn", # encoder decoder
]
MODEL = "meta-llama/Llama-3.2-1B-Instruct"
@pytest.mark.parametrize("model", UNSUPPORTED_MODELS_V1)
def test_reject_unsupported_models(monkeypatch, model):
with monkeypatch.context() as m:
m.setenv("VLLM_USE_V1", "1")
args = AsyncEngineArgs(model=model)
with pytest.raises(NotImplementedError):
_ = args.create_engine_config()
m.delenv("VLLM_USE_V1")
def test_reject_bad_config(monkeypatch):
with monkeypatch.context() as m:
m.setenv("VLLM_USE_V1", "0")
@@ -77,12 +62,6 @@ def test_enable_by_default_fallback(monkeypatch):
assert envs.VLLM_USE_V1
m.delenv("VLLM_USE_V1")
# Should fall back to V0 for supported model.
_ = AsyncEngineArgs(
model=UNSUPPORTED_MODELS_V1[0]).create_engine_config()
assert not envs.VLLM_USE_V1
m.delenv("VLLM_USE_V1")
def test_v1_llm_by_default(monkeypatch):
with monkeypatch.context() as m: