[Rebase] Rebase to vllm v0.18.0 (#2037)

Signed-off-by: tzhouam <tzhouam@connect.ust.hk>
Signed-off-by: Zhou Taichang <tzhouam@connect.ust.hk>
Signed-off-by: gcanlin <canlinguosdu@gmail.com>
Co-authored-by: gcanlin <canlinguosdu@gmail.com>
This commit is contained in:
Zhou Taichangandgcanlin authored and GitHub committed 2026-03-21 20:37:38 +08:00
1 parent 8e12007ac1
commit a90a7690b8
34 files changed
+435 -223

No files matched your search

@@ -24,6 +24,7 @@ os.environ["VLLM_TEST_CLEAN_GPU_MEMORY"] = "1"
# This test is specific to Z-Image LoRA behavior. Keep it focused on a single
# model to reduce runtime and avoid extra downloads.
models = ["Tongyi-MAI/Z-Image-Turbo"]
DIFFUSION_INIT_TIMEOUT_S = 600
@pytest.mark.parametrize("model_name", models)
@@ -76,7 +77,11 @@ def test_diffusion_model(model_name: str, tmp_path: Path):
)
return str(adapter_dir)
m = Omni(model=model_name)
m = Omni(
model=model_name,
stage_init_timeout=DIFFUSION_INIT_TIMEOUT_S,
init_timeout=DIFFUSION_INIT_TIMEOUT_S,
)
try:
# high resolution may cause OOM on L4
height = 256